pith. machine review for the scientific record. sign in

arxiv: 2605.00405 · v1 · submitted 2026-05-01 · 💻 cs.CV

Recognition: unknown

BOLT: Online Lightweight Adaptation for Preparation-Free Heterogeneous Cooperative Perception

Kang Yang , Tianci Bu , Peng Wang , Deying Li , Yongcai Wang
Authors on Pith no claims yet

Pith reviewed 2026-05-09 19:36 UTC · model grok-4.3

classification 💻 cs.CV
keywords cooperative perceptionheterogeneous modelsonline adaptationfeature alignmentego-as-teacher distillationpreparation-free fusionmulti-agent detection
0
0 comments X

The pith

A 0.9M-parameter online adapter lets independently trained detectors fuse features effectively without any prior coordination or labels.

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

The paper addresses the practical barrier that most cooperative perception methods require offline joint training or model-specific adaptation, which cannot happen when agents from different developers meet online. It demonstrates that direct fusion in this preparation-free setting actually hurts performance compared to ego-only detection. BOLT solves this by inserting a lightweight plug-and-play module that treats the ego agent's own high-confidence predictions as a teacher signal to align incoming neighbor features in real time. The module simultaneously lets neighbors supply information in the ego's low-confidence regions. The resulting system consistently beats both unadapted fusion and ego-only baselines on DAIR-V2X and OPV2V across encoder pairs and fusion strategies.

Core claim

BOLT performs online ego-as-teacher distillation to adapt neighboring features into the ego feature domain, using only ego predictions as supervision, so that heterogeneous agents can contribute useful information without ground-truth labels or pre-deployment coordination.

What carries the argument

BOLT module: a small plug-and-play adapter that performs cross-agent feature-domain alignment by distilling from the ego agent's high-confidence predictions while allowing neighbors to fill low-confidence regions.

If this is right

  • Cooperative perception becomes feasible for agents that meet only occasionally and have no shared training history.
  • The same adaptation works across multiple encoder architectures and fusion strategies without retraining the base detectors.
  • Only 0.9 million trainable parameters are needed to obtain up to 32.3 AP@50 improvement over vanilla fusion.
  • Neighbors can contribute information precisely where the ego model is uncertain, without requiring any external labels.

Where Pith is reading between the lines

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

  • The approach could support ad-hoc collaboration among vehicles produced by different manufacturers that never share training data.
  • Similar ego-as-teacher alignment might extend to other multi-agent tasks such as joint mapping or trajectory planning.
  • Deployment on real hardware would need to verify that the online adaptation remains stable under varying latency and bandwidth constraints.

Load-bearing premise

High-confidence ego predictions are accurate and representative enough to serve as a reliable teacher signal for aligning features from other agents.

What would settle it

A controlled test in which ego high-confidence predictions are replaced by random or deliberately mismatched labels, after which BOLT fusion performance falls below the unadapted baseline.

Figures

Figures reproduced from arXiv: 2605.00405 by Deying Li, Kang Yang, Peng Wang, Tianci Bu, Yongcai Wang.

Figure 1
Figure 1. Figure 1: Preparation-free heterogeneous cooperative perception. (a) Prior works; (b) BOLT. cooperative preprocessing. PHCP [29] moves part of the alignment to inference time, but still builds on a cooperatively pre-trained base and warms up its plugin on an unlabeled split of each scene before reporting cooperative performance on the rest. Such a pipeline works well for controlled experiments but fails to meet real… view at source ↗
Figure 2
Figure 2. Figure 2: Effect of BOLT on DAIR-V2X with a LiDAR ego (PointPillars, abbre￾viated PP) and a camera neighbor (LSS￾EfficientNet, abbreviated LSS-E), denoted PP→LSS-E (ego→neighbor) throughout. BOLT improves both feature compatibility (CKA, scale alignment) and detection ac￾curacy (AP@30/50/70). We formalize this practical setting as preparation￾free heterogeneous cooperative perception. Unlike prior methods requiring … view at source ↗
Figure 3
Figure 3. Figure 3: Framework overview. The ego agent’s frozen single-agent path (top) produces teacher [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative BEV results. Top: DAIR-V2X; bottom: OPV2V. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Additional qualitative BEV results for PP [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Additional qualitative BEV results for PP [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Additional qualitative BEV results for SECOND [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Additional qualitative BEV results for SECOND [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Dynamic online convergence of BOLT on DAIR-V2X (PP [PITH_FULL_IMAGE:figures/full_fig_p019_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Precision–recall curves at IoU 0.5 on DAIR-V2X (PP [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗
read the original abstract

Most existing heterogeneous cooperative perception methods depend on prior preparation like offline joint training or tailored collaborator-model adaptation. Such preprocessing is, however, generally impractical in real scenarios, as agents are usually independently trained by different developers and meet occasionally online. This work investigates \emph{preparation-free heterogeneous cooperative perception}, where agents use independently trained single-agent detectors without any pre-deployment coordination. We find direct cross-agent fusion under this setting greatly underperforms ego-only perception. We present BOLT, a lightweight plug-and-play module that adapts neighboring features online via ego-as-teacher distillation, requiring only ego predictions without ground-truth labels. BOLT leverages high-confidence ego perception features to guide cross-agent feature-domain alignment, while enabling neighbors to contribute features in the ego's low-confidence regions. With only 0.9M trainable parameters, BOLT improves AP@50 by up to 32.3 points over vanilla unadapted fusion in the preparation-free setting. It consistently outperforms ego-only results on DAIR-V2X and OPV2V, across different encoder pairs and fusion strategies. Code: https://github.com/sidiangongyuan/BOLT.

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 BOLT, a lightweight (0.9M parameters) plug-and-play module for preparation-free heterogeneous cooperative perception. In this setting, agents use independently trained single-agent detectors with no prior coordination or joint training. BOLT performs online adaptation by treating high-confidence ego predictions as a teacher signal to distill and align features from neighboring agents, allowing neighbors to contribute in ego low-confidence regions. It reports up to +32.3 AP@50 over vanilla unadapted fusion and consistent outperformance of ego-only baselines on DAIR-V2X and OPV2V across encoder pairs and fusion strategies.

Significance. If the quantitative gains prove robust, the work addresses a practically important gap: enabling effective cooperation among heterogeneous, uncoordinated agents without offline preparation. The emphasis on minimal trainable parameters and online-only operation is a clear strength for real-world deployment. The approach builds on standard knowledge-distillation ideas but applies them to a new constraint set; reproducible code is provided, which aids verification.

major comments (2)
  1. [§3, §4] §3 (Method) and §4 (Experiments): The central mechanism relies on ego high-confidence predictions serving as a reliable teacher for cross-agent feature alignment without ground-truth labels. No ablation or analysis is presented on the accuracy of these teacher signals under domain shift, occlusion, or detector-specific biases; if ego predictions contain systematic errors, the distillation could reinforce incorrect alignments rather than transfer useful neighbor information. This assumption is load-bearing for both the fusion improvement and the ego-only outperformance claims.
  2. [§4.2] §4.2 (Quantitative results): The reported AP@50 gains (up to 32.3 points) and outperformance over ego-only are presented without detailed failure-mode analysis, confidence-interval reporting, or controls for post-hoc hyperparameter choices in the online adaptation. The abstract and results sections provide limited experimental details on how high-confidence thresholds are set or how performance varies when ego confidence is low.
minor comments (2)
  1. [§3] Notation for the distillation loss and feature alignment step could be clarified with an explicit equation reference in the main text rather than relying on the supplementary material.
  2. [§4] Figure captions and table footnotes should explicitly state the number of runs or random seeds used to generate the reported means.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We appreciate the referee's recognition of the practical importance of preparation-free heterogeneous cooperative perception and the constructive feedback on validating key assumptions. We address each major comment below and have revised the manuscript to incorporate additional analysis and experimental details.

read point-by-point responses

  1. Referee: [§3, §4] §3 (Method) and §4 (Experiments): The central mechanism relies on ego high-confidence predictions serving as a reliable teacher for cross-agent feature alignment without ground-truth labels. No ablation or analysis is presented on the accuracy of these teacher signals under domain shift, occlusion, or detector-specific biases; if ego predictions contain systematic errors, the distillation could reinforce incorrect alignments rather than transfer useful neighbor information. This assumption is load-bearing for both the fusion improvement and the ego-only outperformance claims.

    Authors: We agree that direct validation of the ego high-confidence predictions as teacher signals is necessary to support the claims. The original manuscript presented performance gains over ego-only baselines as supporting evidence, but we acknowledge this is indirect. In the revised manuscript, we have added a new analysis subsection in §4.3 that quantifies the precision of high-confidence ego predictions (threshold > 0.7) against ground-truth labels under domain shift, varying occlusion levels, and across detector pairs on DAIR-V2X and OPV2V. These diagnostics show precision rates above 88% in the evaluated conditions, indicating limited propagation of systematic errors. We further clarify the selective application of distillation only in high-confidence regions, which allows neighbors to supplement rather than override ego predictions. revision: yes

  2. Referee: [§4.2] §4.2 (Quantitative results): The reported AP@50 gains (up to 32.3 points) and outperformance over ego-only are presented without detailed failure-mode analysis, confidence-interval reporting, or controls for post-hoc hyperparameter choices in the online adaptation. The abstract and results sections provide limited experimental details on how high-confidence thresholds are set or how performance varies when ego confidence is low.

    Authors: We have revised §4.2 to include a failure-mode analysis identifying cases (e.g., extreme multi-agent occlusions) where BOLT provides limited gains, along with mean AP values and standard deviations computed over five random seeds to report confidence intervals. Hyperparameters including the confidence threshold of 0.7 were selected via cross-validation on a held-out validation split prior to test-set evaluation, with no post-hoc tuning on test data; a full sensitivity analysis on the threshold is now provided in the appendix. We also add results stratified by ego confidence levels, showing that BOLT defaults to ego-only features when scene-wide confidence is low and still yields improvements by incorporating neighbor information in mixed-confidence regions. revision: yes

Circularity Check

0 steps flagged

No significant circularity; BOLT applies standard distillation to a new setting without self-referential reductions

full rationale

The paper's core claim rests on an online ego-as-teacher distillation module that aligns neighbor features using high-confidence ego predictions, with reported gains (up to +32.3 AP@50) validated empirically on DAIR-V2X and OPV2V across encoder pairs. No equations define the adaptation loss or alignment in terms of the target performance metric itself, no fitted parameters are relabeled as predictions, and no load-bearing steps reduce to self-citations or prior author ansatzes. The approach is presented as a plug-and-play extension of knowledge-distillation principles to the preparation-free heterogeneous case, with the derivation chain remaining independent of the final empirical outcomes.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Abstract-only review; free parameters and axioms inferred from stated method. No invented entities.

free parameters (1)
  • BOLT module size
    0.9M trainable parameters are introduced and adapted online; their exact initialization or regularization choices are unspecified in the abstract.
axioms (1)
  • domain assumption Ego high-confidence predictions provide a sufficiently accurate teacher signal for feature alignment
    Central to the distillation step; stated implicitly as the basis for guiding neighbor features.

pith-pipeline@v0.9.0 · 5504 in / 1221 out tokens · 39889 ms · 2026-05-09T19:36:35.245858+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

51 extracted references · 9 canonical work pages

  1. [1]

    Cooperative perception for 3d object detection in driving scenarios using infrastructure sensors.IEEE Transactions on Intelligent Transportation Systems, 23(3):1852–1864, 2020

    Eduardo Arnold, Mehrdad Dianati, Robert de Temple, and Saber Fallah. Cooperative perception for 3d object detection in driving scenarios using infrastructure sensors.IEEE Transactions on Intelligent Transportation Systems, 23(3):1852–1864, 2020

    2020

  2. [2]

    F-cooper: Feature based cooperative perception for autonomous vehicle edge computing system using 3d point clouds

    Qi Chen, Xu Ma, Sihai Tang, Jingda Guo, Qing Yang, and Song Fu. F-cooper: Feature based cooperative perception for autonomous vehicle edge computing system using 3d point clouds. InProceedings of the 4th ACM/IEEE Symposium on Edge Computing, pages 88–100, 2019

    2019

  3. [3]

    Cooper: Cooperative perception for connected autonomous vehicles based on 3d point clouds

    Qi Chen, Sihai Tang, Qing Yang, and Song Fu. Cooper: Cooperative perception for connected autonomous vehicles based on 3d point clouds. In2019 IEEE 39th International Conference on Distributed Computing Systems (ICDCS), pages 514–524. IEEE, 2019

    2019

  4. [4]

    Transiff: An instance-level feature fusion framework for vehicle-infrastructure cooperative 3d detection with transformers

    Ziming Chen, Yifeng Shi, and Jinrang Jia. Transiff: An instance-level feature fusion framework for vehicle-infrastructure cooperative 3d detection with transformers. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 18205–18214, 2023

    2023

  5. [5]

    Negocollab: A common representation negotiation approach for heteroge- neous collaborative perception

    Shao Congzhang, Quan Yuan, Guiyang Luo, Yue Hu, Danni Wang, Liu Yilin, Rui Pan, Bo Chen, and Jinglin Li. Negocollab: A common representation negotiation approach for heteroge- neous collaborative perception. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems

  6. [6]

    The pascal visual object classes (voc) challenge.International journal of computer vision, 88 (2):303–338, 2010

    Mark Everingham, Luc Van Gool, Christopher KI Williams, John Winn, and Andrew Zisserman. The pascal visual object classes (voc) challenge.International journal of computer vision, 88 (2):303–338, 2010

    2010

  7. [7]

    Quest: Query stream for practical cooperative perception

    Siqi Fan, Haibao Yu, Wenxian Yang, Jirui Yuan, and Zaiqing Nie. Quest: Query stream for practical cooperative perception. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 18436–18442. IEEE, 2024

    2024

  8. [8]

    Stamp: Scalable task and model-agnostic collaborative perception.arXiv preprint arXiv:2501.18616, 2025

    Xiangbo Gao, Runsheng Xu, Jiachen Li, Ziran Wang, Zhiwen Fan, and Zhengzhong Tu. Stamp: Scalable task and model-agnostic collaborative perception.arXiv preprint arXiv:2501.18616, 2025

    work page arXiv 2025

  9. [9]

    Yi Guo and Jiaqi Ma. Leveraging existing high-occupancy vehicle lanes for mixed-autonomy traffic management with emerging connected automated vehicle applications.Transportmetrica A: Transport Science, 16(3):1375–1399, 2020

    2020

  10. [10]

    Collabora- tive perception in autonomous driving: Methods, datasets, and challenges.IEEE Intelligent Transportation Systems Magazine, 15(6):131–151, 2023

    Yushan Han, Hui Zhang, Huifang Li, Yi Jin, Congyan Lang, and Yidong Li. Collabora- tive perception in autonomous driving: Methods, datasets, and challenges.IEEE Intelligent Transportation Systems Magazine, 15(6):131–151, 2023

    2023

  11. [11]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

    2016

  12. [12]

    Momentum contrast for unsupervised visual representation learning

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual representation learning. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9729–9738, 2020. 10

    2020

  13. [13]

    Parameter-efficient transfer learning for NLP

    Neil Houlsby, Andrei Giurgiu, Stanislaw Jastrzebski, Bruna Morrone, Quentin de Laroussilhe, Andrea Gesmundo, Mona Attariyan, and Sylvain Gelly. Parameter-efficient transfer learning for NLP. InICML, 2019

    2019

  14. [14]

    LoRA: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-rank adaptation of large language models. InICLR, 2022

    2022

  15. [15]

    Test-time learning for large language models.arXiv preprint arXiv:2505.20633, 2025

    Jinwu Hu, Zhitian Zhang, Guohao Chen, Xutao Wen, Chao Shuai, Wei Luo, Bin Xiao, Yuanqing Li, and Mingkui Tan. Test-time learning for large language models.CoRR, abs/2505.20633,

    work page arXiv

  16. [16]

    doi: 10.48550/arXiv.2505.20633

    work page doi:10.48550/arxiv.2505.20633

  17. [17]

    Where2comm: Communication-efficient collaborative perception via spatial confidence maps.Advances in neural information processing systems, 35:4874–4886, 2022

    Yue Hu, Shaoheng Fang, Zixing Lei, Yiqi Zhong, and Siheng Chen. Where2comm: Communication-efficient collaborative perception via spatial confidence maps.Advances in neural information processing systems, 35:4874–4886, 2022

    2022

  18. [18]

    Communication-efficient collaborative perception via information filling with codebook

    Yue Hu, Juntong Peng, Sifei Liu, Junhao Ge, Si Liu, and Siheng Chen. Communication-efficient collaborative perception via information filling with codebook. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15481–15490, 2024

    2024

  19. [19]

    Arbitrary style transfer in real-time with adaptive instance normalization

    Xun Huang and Serge Belongie. Arbitrary style transfer in real-time with adaptive instance normalization. InProceedings of the IEEE International Conference on Computer Vision, pages 1501–1510, 2017

    2017

  20. [20]

    Roco: Robust cooperative perception by iterative object matching and pose adjustment

    Zhe Huang, Shuo Wang, Yongcai Wang, Wanting Li, Deying Li, and Lei Wang. Roco: Robust cooperative perception by iterative object matching and pose adjustment. InACM Multimedia 2024, 2024

    2024

  21. [21]

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

    Alex H. Lang, Sourabh V ora, Holger Caesar, Lubing Zhou, Jiong Yang, and Oscar Beijbom. Pointpillars: Fast encoders for object detection from point clouds. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2019

    2019

  22. [22]

    Latency-aware collaborative perception

    Zixing Lei, Shunli Ren, Yue Hu, Wenjun Zhang, and Siheng Chen. Latency-aware collaborative perception. InComputer Vision – ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part XXXII, page 316–332, Berlin, Heidelberg, 2022. Springer-Verlag. ISBN 978-3-031-19823-6. doi: 10.1007/978-3-031-19824-3_19. URL https://doi.o...

    work page doi:10.1007/978-3-031-19824-3_19 2022

  23. [23]

    Focal loss for dense object detection

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. Focal loss for dense object detection. InProceedings of the IEEE international conference on computer vision, pages 2980–2988, 2017

    2017

  24. [24]

    Linking modality isolation in heterogeneous collaborative perception.arXiv preprint arXiv:2603.00609, 2026

    Changxing Liu, Zichen Chao, and Siheng Chen. Linking modality isolation in heterogeneous collaborative perception.arXiv preprint arXiv:2603.00609, 2026

    work page arXiv 2026

  25. [25]

    Fusioneye: Perception sharing for connected vehicles and its bandwidth-accuracy trade-offs

    Hansi Liu, Pengfei Ren, Shubham Jain, Mohannad Murad, Marco Gruteser, and Fan Bai. Fusioneye: Perception sharing for connected vehicles and its bandwidth-accuracy trade-offs. In 2019 16th Annual IEEE International Conference on Sensing, Communication, and Networking (SECON), pages 1–9. IEEE, 2019

    2019

  26. [26]

    Ttt++: When does self-supervised test-time training fail or thrive? In Advances in Neural Information Processing Systems, 2021

    Yuejiang Liu, Parth Kothari, Bastien van Delft, Baptiste Bellot-Gurlet, Taylor Mordan, and Alexandre Alahi. Ttt++: When does self-supervised test-time training fail or thrive? In Advances in Neural Information Processing Systems, 2021

    2021

  27. [27]

    Robust collaborative 3d object detection in presence of pose errors

    Yifan Lu, Quanhao Li, Baoan Liu, Mehrdad Dianati, Chen Feng, Siheng Chen, and Yanfeng Wang. Robust collaborative 3d object detection in presence of pose errors. In2023 IEEE International Conference on Robotics and Automation (ICRA), pages 4812–4818. IEEE, 2023

    2023

  28. [28]

    An extensible framework for open heterogeneous collaborative perception.arXiv preprint arXiv:2401.13964, 2024

    Yifan Lu, Yue Hu, Yiqi Zhong, Dequan Wang, Siheng Chen, and Yanfeng Wang. An extensible framework for open heterogeneous collaborative perception.arXiv preprint arXiv:2401.13964, 2024

    work page arXiv 2024

  29. [29]

    Lift, splat, shoot: Encoding images from arbitrary camera rigs by implicitly unprojecting to 3d, 2020

    Jonah Philion and Sanja Fidler. Lift, splat, shoot: Encoding images from arbitrary camera rigs by implicitly unprojecting to 3d, 2020. 11

    2020

  30. [30]

    You share beliefs, i adapt: Progressive hetero- geneous collaborative perception

    Hao Si, Ehsan Javanmardi, and Manabu Tsukada. You share beliefs, i adapt: Progressive hetero- geneous collaborative perception. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 27521–27530, 2025

    2025

  31. [31]

    Traf-align: Trajectory-aware feature alignment for asynchronous multi-agent perception

    Zhiying Song, Lei Yang, Fuxi Wen, and Jun Li. Traf-align: Trajectory-aware feature alignment for asynchronous multi-agent perception. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 12048–12057, 2025

    2025

  32. [32]

    Collaborative multi-object tracking with conformal uncertainty propagation.IEEE Robotics and Automation Letters, 9(4):3323–3330, 2024

    Sanbao Su, Songyang Han, Yiming Li, Zhili Zhang, Chen Feng, Caiwen Ding, and Fei Miao. Collaborative multi-object tracking with conformal uncertainty propagation.IEEE Robotics and Automation Letters, 9(4):3323–3330, 2024

    2024

  33. [33]

    Test-time training with self-supervision for generalization under distribution shifts

    Yu Sun, Xiaolong Wang, Zhuang Liu, John Miller, Alexei A Efros, and Moritz Haus. Test-time training with self-supervision for generalization under distribution shifts. InInternational Conference on Machine Learning, 2020

    2020

  34. [34]

    Efficientnet: Rethinking model scaling for convolutional neural networks

    Mingxing Tan and Quoc Le. Efficientnet: Rethinking model scaling for convolutional neural networks. InInternational conference on machine learning, pages 6105–6114. PMLR, 2019

    2019

  35. [35]

    Tent: Fully test-time adaptation by entropy minimization

    Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Olshausen, and Trevor Darrell. Tent: Fully test-time adaptation by entropy minimization. InInternational Conference on Learning Representations, 2021

    2021

  36. [36]

    V2vnet: Vehicle-to-vehicle communication for joint perception and prediction

    Tsun-Hsuan Wang, Sivabalan Manivasagam, Ming Liang, Bin Yang, Wenyuan Zeng, and Raquel Urtasun. V2vnet: Vehicle-to-vehicle communication for joint perception and prediction. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part II 16, pages 605–621. Springer, 2020

    2020

  37. [37]

    Asynchrony- robust collaborative perception via bird’s eye view flow.Advances in Neural Information Processing Systems, 36:28462–28477, 2023

    Sizhe Wei, Yuxi Wei, Yue Hu, Yifan Lu, Yiqi Zhong, Siheng Chen, and Ya Zhang. Asynchrony- robust collaborative perception via bird’s eye view flow.Advances in Neural Information Processing Systems, 36:28462–28477, 2023

    2023

  38. [38]

    One is plenty: A polymorphic feature interpreter for immutable heterogeneous collaborative perception

    Yuchen Xia, Quan Yuan, Guiyang Luo, Xiaoyuan Fu, Yang Li, Xuanhan Zhu, Tianyou Luo, Siheng Chen, and Jinglin Li. One is plenty: A polymorphic feature interpreter for immutable heterogeneous collaborative perception. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 1592–1601, 2025

    2025

  39. [39]

    Hm-vit: Hetero-modal vehicle-to-vehicle cooperative perception with vision transformer

    Hao Xiang, Runsheng Xu, and Jiaqi Ma. Hm-vit: Hetero-modal vehicle-to-vehicle cooperative perception with vision transformer. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 284–295, 2023

    2023

  40. [40]

    V2x-real: a largs-scale dataset for vehicle-to-everything cooperative perception

    Hao Xiang, Zhaoliang Zheng, Xin Xia, Runsheng Xu, Letian Gao, Zewei Zhou, Xu Han, Xinkai Ji, Mingxi Li, Zonglin Meng, et al. V2x-real: a largs-scale dataset for vehicle-to-everything cooperative perception. InEuropean Conference on Computer Vision, pages 455–470. Springer, 2024

    2024

  41. [41]

    Cobevt: Cooperative bird’s eye view semantic segmentation with sparse transformers, 2022

    Runsheng Xu, Zhengzhong Tu, Hao Xiang, Wei Shao, Bolei Zhou, and Jiaqi Ma. Cobevt: Cooperative bird’s eye view semantic segmentation with sparse transformers, 2022

    2022

  42. [42]

    Opv2v: An open benchmark dataset and fusion pipeline for perception with vehicle-to-vehicle communication

    Runsheng Xu, Hao Xiang, Xin Xia, Xu Han, Jinlong Li, and Jiaqi Ma. Opv2v: An open benchmark dataset and fusion pipeline for perception with vehicle-to-vehicle communication. In2022 International Conference on Robotics and Automation (ICRA), pages 2583–2589. IEEE, 2022

    2022

  43. [43]

    V2v4real: A real-world large-scale dataset for vehicle-to-vehicle cooperative perception

    Runsheng Xu, Xin Xia, Jinlong Li, Hanzhao Li, Shuo Zhang, Zhengzhong Tu, Zonglin Meng, Hao Xiang, Xiaoyu Dong, Rui Song, et al. V2v4real: A real-world large-scale dataset for vehicle-to-vehicle cooperative perception. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13712–13722, 2023

    2023

  44. [44]

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

    Yan Yan, Yuxing Mao, and Bo Li. Second: Sparsely embedded convolutional detection.Sensors, 18(10):3337, 2018. 12

    2018

  45. [45]

    How2comm: Communication-efficient and collaboration-pragmatic multi-agent perception.Advances in Neural Information Processing Systems, 36:25151–25164, 2023

    Dingkang Yang, Kun Yang, Yuzheng Wang, Jing Liu, Zhi Xu, Rongbin Yin, Peng Zhai, and Lihua Zhang. How2comm: Communication-efficient and collaboration-pragmatic multi-agent perception.Advances in Neural Information Processing Systems, 36:25151–25164, 2023

    2023

  46. [46]

    Safe multi-agent navigation guided by goal- conditioned safe reinforcement learning

    Kang Yang, Tianci Bu, Lantao Li, Chunxu Li, Yongcai Wang, and Deying Li. Is discretization fusion all you need for collaborative perception? In2025 IEEE International Conference on Robotics and Automation (ICRA), pages 9590–9596, 2025. doi: 10.1109/ICRA55743.2025. 11128776

    work page doi:10.1109/icra55743.2025 2025

  47. [47]

    Eimc: Efficient instance-aware multi-modal collaborative perception.arXiv preprint arXiv:2603.02532, 2026

    Kang Yang, Peng Wang, Lantao Li, Tianci Bu, Chen Sun, Deying Li, and Yongcai Wang. Eimc: Efficient instance-aware multi-modal collaborative perception.arXiv preprint arXiv:2603.02532, 2026

    work page arXiv 2026

  48. [48]

    Dual test-time training for out-of-distribution recommender system.IEEE Transactions on Knowledge and Data Engineering, 37(6):3312–3326, 2025

    Xihong Yang, Yiqi Wang, Jin Chen, Wenqi Fan, Xiangyu Zhao, En Zhu, Xinwang Liu, and Defu Lian. Dual test-time training for out-of-distribution recommender system.IEEE Transactions on Knowledge and Data Engineering, 37(6):3312–3326, 2025. doi: 10.1109/TKDE.2025.3548160

    work page doi:10.1109/tkde.2025.3548160 2025

  49. [50]

    Dair-v2x: A large-scale dataset for vehicle-infrastructure cooperative 3d object detection

    Haibao Yu, Yizhen Luo, Mao Shu, Yiyi Huo, Zebang Yang, Yifeng Shi, Zhenglong Guo, Hanyu Li, Xing Hu, Jirui Yuan, et al. Dair-v2x: A large-scale dataset for vehicle-infrastructure cooperative 3d object detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21361–21370, 2022

    2022

  50. [51]

    Sparsealign: A fully sparse framework for cooperative object detection

    Yunshuang Yuan, Yan Xia, Daniel Cremers, and Monika Sester. Sparsealign: A fully sparse framework for cooperative object detection. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 22296–22305, 2025. 13 A Reproducibility and Evaluation Protocol Encoder architectures.We use three independently pre-trained encoders, all produci...

    2025

  51. [52]

    We view this latency–convergence trade-off as a deployment tuning knob rather than a fundamental blocker

    In deployment scenarios with tight real-time budgets, this cost may be partially amortized by updating only at a lower rate (e.g., every k samples) rather than every frame, at the expense of slightly slower convergence. We view this latency–convergence trade-off as a deployment tuning knob rather than a fundamental blocker. 20 Weak ego encoder.BOLT relies...