Pith. sign in

REVIEW 3 major objections 5 minor 35 references

HeteroPROMPT shows that heterogeneous collaborative perception reduces to a per-modality prompt-tuning problem, alignable with 0.064M parameters and no metadata sharing.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

A prompt-based, privacy-preserving heterogeneity-alignment framework that uses low-rank FiLM to adapt BEV features and an autoencoder-based classifier for metadata-free modality routing.

T0 review reviewed 2026-08-01 challenge →

load-bearing objection HeteroPROMPT is a genuinely parameter-efficient prompt-based heterogeneous CP method, but its SOTA claim is undermined by omitting the two closest baselines: the authors' own Faster-HEAL and PEARL. the 3 major comments →

arxiv 2607.26283 v1 pith:VZGJPBIW submitted 2026-07-28 cs.CV cs.RO

HeteroPROPMT: A Real-time and Privacy-Preserving Heterogeneous Collaborative Perception Framework

Armin Maleki , Hayder Radha This is my paper
classification cs.CV cs.RO
keywords collaborative perceptionheterogeneous agentsfeature alignmentvisual promptsFiLMlow-rank decompositionprivacy preservationmodality classification
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

The paper claims that heterogeneous collaborative perception—where vehicles with different sensors, backbones, and training data share intermediate features—can be made practical and privacy-preserving without retraining the heavy fusion and detection stacks. Its central proposal, HeteroPROMPT, adapts each newly joining agent by training only two tiny 'prompt' modules (DRIFT) that modulate received Bird's Eye View features through rank-reduced FiLM affine transforms. With only 0.064M trainable parameters, it reports higher detection Average Precision than state-of-the-art retraining-based and interpreter-based methods on two standard datasets. To avoid exposing proprietary sensor/model metadata, an autoencoder compresses the received features into a compact latent space and a lightweight classifier routes each agent to the correct DRIFT pair, achieving >99.99% modality classification accuracy on the evaluated closed set. A sympathetic reader should care because this is the first time the heterogeneous-collaboration problem is reduced to a per-modality prompt-tuning task with near-zero adaptation cost.

Core claim

HeteroPROMPT establishes that feature-space shifts between heterogeneous agents can be closed by two low-rank visual-prompt-conditioned FiLM modules placed before and after the feature resizer, while keeping the ego agent's fusion/head and the joining agent's encoder frozen. The prompts are decomposed via PARAFAC into factor matrices of rank R, cutting trainable parameters from C×H×W to R(C+H+W), which is what makes the whole system fit in 0.064M parameters. A second, deployment-time discovery is that modality identity is recoverable from compact features: after Stage-1 training, an autoencoder learned on ego-domain features, together with a three-layer MLP on the pooled latent, classifies t

What carries the argument

The load-bearing object is DRIFT (Decomposed-Rank Interpreter for FiLM Tuning), a per-modality module that computes channel-wise affine parameters (gamma, beta) from a learnable visual prompt and applies them to BEV features. The prompt itself is stored as a PARAFAC low-rank decomposition (three factor matrices over channel, height, width), which is what collapses parameter count. Two DRIFT modules are inserted at different pipeline depths: one on the raw communicated feature (matching the sender's resolution) and one after the resizer (matching the ego's unified resolution), so alignment happens both in the sender's domain and in the unified domain. The second key mechanism is the autoencod

Load-bearing premise

The deployment pipeline assumes the joining agent's sensor/encoder/training combination is one of the closed set of modalities used to train DRIFT and the classifier (m2–m9); an unseen combination gets routed to a wrong interpreter and alignment fails.

What would settle it

Take any perception encoder not in the paper's modality set (for instance, a transformer-based LiDAR detector or a camera model trained on a different city), feed its BEV features to the classifier, and check whether routing picks the correct DRIFT and whether AP50 stays within the reported range. If an unseen-but-similar modality is misrouted with high confidence, the metadata-free routing claim does not generalize outside the closed set.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • New agents can be added by training only two low-rank prompt modules per modality, so fleet expansion becomes cheap and does not require access to the agent's encoder weights.
  • Because routing uses features, not metadata, collaboration can proceed without partners revealing their sensor/model configuration, reducing privacy exposure and attack surface.
  • The >99.99% modality classification means reliable module selection is feasible at deployment latencies, not just in principle.
  • The 94–99.6% parameter reduction relative to retraining baselines suggests the approach fits on onboard compute for real-time inference.
  • If the frozen single-agent stack is preserved, agents can fall back to their own perception when collaboration is interrupted without any retraining.

Where Pith is reading between the lines

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

  • The closed-set assumption is the main boundary: the paper tests only modalities m2–m9; an open-set variant could treat the classifier's softmax confidence as a rejection rule and fall back to no collaboration or a generic aligner for unknown modalities.
  • The compact AE latent space might serve as a general domain fingerprint for other tasks (e.g., predicting pose noise, latency, or best fusion weight) not just modality routing.
  • One could test whether the DRIFT alignment transfers to entirely different downstream heads (e.g., segmentation) without retraining, since alignment happens in a task-agnostic unified feature space.
  • The method implicitly assumes feature distributions are stable for a given modality; a natural extension is to learn a DRIFT per training-domain per sensor rather than per sensor+model, which would shrink storage further.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes HeteroPROMPT, a framework for heterogeneous collaborative perception. It aligns BEV features from newly joining agents to an ego-centric unified space using two lightweight prompt-conditioned FiLM modules (DRIFT) with PARAFAC low-rank prompts, while freezing agent encoders and the ego fusion/detection stack. A second contribution is metadata-free modality routing: an autoencoder compresses communicated BEV features to a compact latent space, and a lightweight MLP classifier predicts the joining agent's modality to select the appropriate DRIFT pair. Experiments on OPV2V-H and V2XSet compare against HEAL, CoBEVT, and HM-ViT, reporting higher AP50/AP70 with far fewer trainable parameters, plus a claimed >99.99% modality-classification accuracy. The central idea is plausible and the component-level ablations support the design, but the evaluation omits the two most closely related prompt-based baselines, and the stated classification accuracy is quantitatively inconsistent with the reported confusion matrix.

Significance. If the claims hold, HeteroPROMPT would be a useful step toward practical heterogeneous collaborative perception: parameter-efficient adaptation that avoids retraining large perception stacks, preserves agent encoder privacy, and enables metadata-free routing. The PARAFAC decomposition of visual prompts is a sensible way to reduce adaptation cost, and the two-point FiLM placement with ablations is a reasonable engineering contribution. The use of public datasets (OPV2V-H, V2XSet) and the inclusion of training-throughput/memory measurements are strengths. However, the paper's 'state-of-the-art' claim currently rests on a comparison set that excludes the two prior prompt-based heterogeneous-CP methods it identifies as most related, and the >99.99% classification accuracy cannot be reconciled with the one-error confusion matrix under the stated test-set sizes. The core approach is defensible, but these issues require substantive experimental or textual revision.

major comments (3)
  1. [II-B, Table II] The related-work section identifies Faster-HEAL [16] and PEARL [17] as prompt-based, parameter-efficient, privacy-preserving heterogeneous CP methods that are directly comparable to HeteroPROMPT, yet neither appears in Table II. As a result, the central claim of 'state-of-the-art' AP and 'orders-of-magnitude fewer parameters' is not tested against the closest baselines. Please add quantitative comparisons to [16] and [17] on the same settings, or justify their exclusion with concrete numbers; otherwise the SOTA assertion is unsupported by the presented evidence.
  2. [IV-C, Fig. 3 vs Abstract/V] The paper states that the modality classifier achieves '>99.99%' accuracy, with 'one confusion between m5 and m2.' Section IV-C reports per-modality test samples of 30 (or 140/30/30 splits) for each modality. With 8 modalities and 30 test samples each, one error yields 99.58%; even with 140 test samples per modality, one error yields 99.11%. Both are far below 99.99%. The abstract and conclusion repeat the >99.99% claim. Please report the exact total number of test samples and the true accuracy, and adjust all occurrences of the claim accordingly.
  3. [Table II, IV-B] The reported AP improvements over HEAL are small (+1.0 to +1.8 points AP50), and no error bars, confidence intervals, or multiple seeds are reported. Without such variability information, the claim of 'consistently improves' is not statistically grounded, especially when the differences are comparable to run-to-run noise. Please add results over at least three seeds (mean and std), or otherwise temper the claim to a single-run observation.
minor comments (5)
  1. [Table II, Abstract] The abstract and introduction say 'orders of magnitude fewer trainable parameters,' citing 0.064M vs 15M. However, Table II lists HEAL parameters as 15.0M for +m2, 1.1M for +m3, and 1.9M for +m4. For +m3 and +m4 the reduction is a factor of 17–30, i.e., about one order of magnitude, not 'orders of magnitude.' Please either explain what the 15M/1.1M/1.9M counts represent or adjust the wording.
  2. [Contributions, Table III] The contribution bullet claims 'reducing average training overhead by 96%.' Table III reports 2.5x higher throughput, 2.27x higher TFLOPs/s, and 45.7% lower peak GPU memory — none of which directly translates to a 96% training-overhead reduction. Presumably this refers to trainable parameter reduction, not wall-clock or memory overhead. Please clarify the metric used for the 96% claim.
  3. [III-C, V] The paper correctly acknowledges in Section III-C that Stage 2 adaptation is performed offline for a closed set of common sensor/model configurations, and Section V defers open-set generalization to future work. This is an important limitation; please state it explicitly in the abstract or introduction so that the 'real-time deployment' framing is not overstated.
  4. [IV-C, Fig. 2] The t-SNE visualization is qualitative. Consider reporting a quantitative separation metric (e.g., k-NN accuracy or adjusted Rand index on the latent space) to support the claim that the compact space 'forms distinct modality clusters.'
  5. [I, IV-B] The 'privacy-preserving' claim is tied to not sharing sensor/model metadata, but intermediate BEV features themselves may expose scene information. A sentence acknowledging that privacy is defined relative to metadata disclosure, not feature content, would improve clarity.

Circularity Check

0 steps flagged

No construction-level circularity: the central AP results are measured against external baselines on public datasets. The self-cited Faster-HEAL and PEARL baselines are omitted from Table II, a benchmarking gap rather than a circular derivation; the >99.99% classifier claim is statistically unsupported but not circular.

full rationale

The derivation chain is self-contained. Stage 1 follows the external HEAL recipe to learn an ego-centric unified space; Stage 2 trains modality-specific DRIFT modules (FiLM + PARAFAC prompts) on training data of each modality while freezing encoders and the fusion/head; Stage 1.5 trains an AutoEncoder on ego features; and the modality classifier is a supervised MLP on AE latent codes. The main quantitative claims in Tables II and III are measured against external baselines (HEAL, CoBEVT, HM-ViT) on public OPV2V-H and V2XSet data, so they are not equivalent to the paper's inputs by construction. The self-citations to Faster-HEAL [16] and PEARL [17] in Section II-B are related-work descriptions, not load-bearing derivations; however, their absence from the experimental comparison weakens the 'state-of-the-art' claim, which is a benchmarking/completeness issue rather than circularity. The >99.99% modality-classifier accuracy (Section IV-C, Fig. 3) is statistically implausible with only 30 test samples per modality and one reported confusion, and the system is explicitly closed-set (Section III-C: 'Stage 2 adaptation is performed offline for a set of common sensor/model configurations'), but these are evidence-limitation concerns. No equation is fitted then renamed as a prediction: the classifier is evaluated on held-out features and DRIFT modules are trained per modality with frozen stacks. Score 1 reflects only the presence of minor non-load-bearing self-citations.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central claim rests on architectural choices (PARAFAC rank, AE latent size, FiLM width) that are tuned on held-out data or ablation, and on domain assumptions about the sufficiency of FiLM-style modulation and the closed-world nature of deployment modalities. No new physical entities are introduced.

free parameters (4)
  • PARAFAC rank R = 16
    Prompt factor rank chosen via ablation (Fig. 4) as the best accuracy–complexity trade-off; central to prompt parameter count.
  • AutoEncoder latent size = (64, 8, 16)
    Selected as the lowest test loss in Table IV; used for modality classification and routing. Selection on the test set is a mild form of data-dependent tuning.
  • FiLM CNN hidden channels = 32
    Chosen via ablation in Fig. 4 as the best trade-off for R=16; affects the FiLM module capacity.
  • Loss weight λ (reconstruction vs detection)
    Balances the Stage 1.5 objective in Eq. (4); no value is specified in the text.
axioms (5)
  • domain assumption BEV features from different agent modalities can be aligned by channel- and spatial-wise affine modulation (FiLM) alone, without modifying encoders or detectors
    Core premise of Stage 2; Section III-B and Eq. (2).
  • domain assumption A compact autoencoder latent space of ego BEV features preserves enough modality-discriminative information for a classifier to identify a joining agent's modality
    Relies on Stage 1.5 AE and the MLP classifier; Section III-C.
  • domain assumption The set of possible modalities is known and finite; DRIFT modules and classifier are trained only for a fixed set of common sensor/model configurations
    Section III-C: 'Stage 2 adaptation is performed offline for a set of common sensor/model configurations' — the closed-world assumption underlies the routing scheme.
  • standard math PARAFAC rank-R decomposition (Eq. 3) is a valid low-rank approximation of the full prompt tensor
    Standard tensor decomposition; assumed to retain alignment quality, supported only by ablations.
  • domain assumption The ego-centric unified feature space learned in Stage 1 (following HEAL) is a suitable target space for heterogeneous collaboration
    Stage 1 adopts the homogeneous base of HEAL [12]; no independent verification that this space is optimal for all downstream heterogeneous alignment.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of HeteroPROPMT: A Real-time and Privacy-Preserving Heterogeneous Collaborative Perception Framework." pith.science (2026). https://pith.science/paper/VZGJPBIW

@misc{pith2026260726283,
  author       = {Pith},
  title        = {Pith review of: HeteroPROPMT: A Real-time and Privacy-Preserving Heterogeneous Collaborative Perception Framework},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VZGJPBIW}},
  note         = {Machine review of arXiv:2607.26283}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Collaborative Perception (CP) improves autonomous systems' awareness of their surroundings by sharing sensor data, intermediate features, and detection results. In real-world deployments, however, collaborating vehicles often use heterogeneous sensors, perception models, datasets, and training domains, creating feature-space shifts that degrade downstream fusion and detection. Existing approaches typically retrain fusion and detection components or introduce modality-specific feature interpreters. These methods scale poorly to newly joining agents and often require access to proprietary metadata, raising privacy concerns. We propose HeteroPROMPT, a real-time and privacy-preserving framework for heterogeneous collaborative perception. HeteroPROMPT rapidly aligns each heterogeneous agent's features with an ego-centric unified feature space through modular prompts and lightweight learning-based tuning, while keeping agent encoders and the collaborative fusion and detection stacks frozen. Its visual prompt-based training and inference modulate Bird's Eye View (BEV) features across channels and spatial locations with low computational overhead. For metadata-free deployment, an autoencoder learns a compact unified representation and extracts modality cues from shared features, enabling real-time modality classification and routing to the appropriate HeteroPROMPT modules without exposing proprietary agent information. Experiments on the OPV2V-H and V2XSet datasets show that HeteroPROMPT improves Average Precision over state-of-the-art heterogeneous CP methods while using orders of magnitude fewer trainable parameters. This offers a scalable and practical CP solution. The proposed modality classifier also predicts the joining agent's modality from compact features with greater than 99.99 percent accuracy during deployment. Code will be available at https://github.com/arminmaleki007/HeteroPROMPT.

Figures

Figures reproduced from arXiv: 2607.26283 by Armin Maleki, Hayder Radha.

Figure 1
Figure 1. Figure 1: Overview of HETEROPROMPT. (a) Stage 1 learns an ego-centric unified BEV feature space for collaborative fusion and detection. (b) Stage 1.5 trains an AE to obtain a compact latent representation from ego-domain BEV features. (c) Color/marker legend. (d) Stage 2 fine-tunes two DRIFT modules to align a newly joining heterogeneous agent to the unified space. (e) DRIFT uses PARAFAC low-rank prompt-conditioned … view at source ↗
Figure 2
Figure 2. Figure 2: t-SNE visualization of AE compact features (flattened) for modality [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Test confusion matrix for deployment modality classification using [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Accuracy-complexity ablations for Stage 2 new-agent adaptation [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

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

35 extracted references · 3 linked inside Pith

  1. [1]

    Hao et al., Research challenges and progress in the end-to-end v2x cooperative autonomous driving competition

    R. Hao et al., Research challenges and progress in the end-to-end v2x cooperative autonomous driving competition. In Proceedings of the IEEE/CVF International Conference on Computer Vision 2025 (pp. 1828-1839)

    2025

  2. [2]

    Yang et al., How2comm: Communication-efficient and collaboration-pragmatic multi-agent perception

    D. Yang et al., How2comm: Communication-efficient and collaboration-pragmatic multi-agent perception. Advances in Neural Information Processing Systems. 2023 Dec 15

    2023

  3. [3]

    J. Xu, Y . Zhang, Z. Cai, D. Huang. Cosdh: communication- efficient collaborative perception via supply-demand awareness and intermediate-late hybridization. In Proceedings of the Computer Vision and Pattern Recognition Conference 2025 (pp. 6834-6843)

    2025

  4. [4]

    Wei et al., Asynchrony-robust collaborative perception via bird’s eye view flow

    S. Wei et al., Asynchrony-robust collaborative perception via bird’s eye view flow. Advances in Neural Information Processing Systems. 2023 Dec 15;36:28462-77

    2023

  5. [5]

    J. Wang, T. Nordstr ¨om. Latency robust cooperative perception using asynchronous feature fusion. In 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (W ACV) 2025 Feb 26 (pp. 1-10)

    2025

  6. [6]

    Huang et al., Roco: Robust cooperative perception by iterative object matching and pose adjustment

    Z. Huang et al., Roco: Robust cooperative perception by iterative object matching and pose adjustment. In Proceedings of the 32nd ACM International Conference on Multimedia 2024 Oct 28 (pp. 7833-7842)

    2024

  7. [7]

    J. Xu, C. Song, C. Shi, H. Liu, Q. Wang. UncertainBEV: Uncertainty- aware BEV fusion for roadside 3D object detection. Image and Vision Computing. 2025 Jun 1;159:105567

    2025

  8. [8]

    Zhi et al., Cross-domain generalization for lidar-based 3d object detection in infrastructure and vehicle environments

    P. Zhi et al., Cross-domain generalization for lidar-based 3d object detection in infrastructure and vehicle environments. Sensors. 2025 Jan 27;25(3):767

    2025

  9. [9]

    HM-ViT: Hetero-modal vehicle-to vehi- cle cooperative perception with vision transformer,

    H. Xiang, R. Xu, and J. Ma, “HM-ViT: Hetero-modal vehicle-to vehi- cle cooperative perception with vision transformer,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), Paris, France, 2023, pp. 284–295

    2023

  10. [10]

    Zhao et al., HeatV2X: Scalable Heterogeneous Collaborative Perception via Efficient Alignment and Interaction

    Y . Zhao et al., HeatV2X: Scalable Heterogeneous Collaborative Perception via Efficient Alignment and Interaction. arXiv preprint arXiv:2511.10211. 2025 Nov 13

    arXiv 2025

  11. [11]

    Shao et al., NegoCollab: A Common Representation Negotiation Approach for Heterogeneous Collaborative Perception

    C. Shao et al., NegoCollab: A Common Representation Negotiation Approach for Heterogeneous Collaborative Perception. arXiv preprint arXiv:2510.27647. 2025 Oct 31

    arXiv 2025

  12. [12]

    Lu et al., An extensible framework for open heterogeneous collab- orative perception

    Y . Lu et al., An extensible framework for open heterogeneous collab- orative perception. arXiv preprint arXiv:2401.13964. January 2024

    Pith/arXiv arXiv 2024

  13. [13]

    Xin et al., PnPDA+: A Meta Feature-Guided Domain Adapter for Collaborative Perception

    L. Xin et al., PnPDA+: A Meta Feature-Guided Domain Adapter for Collaborative Perception. World Electric Vehicle Journal. 2025 Jun 21;16(7):343

    2025

  14. [14]

    Xia et al., One is Plenty: A Polymorphic Feature Interpreter for Immutable Heterogeneous Collaborative Perception

    Y . Xia et al., One is Plenty: A Polymorphic Feature Interpreter for Immutable Heterogeneous Collaborative Perception. In Proceedings of the Computer Vision and Pattern Recognition Conference 2025 (pp. 1592-1601)

    2025

  15. [15]

    Zhou et al., Pragmatic Heterogeneous Collaborative Percep- tion via Generative Communication Mechanism

    J. Zhou et al., Pragmatic Heterogeneous Collaborative Percep- tion via Generative Communication Mechanism. arXiv preprint arXiv:2510.19618. 2025 Oct 22

    arXiv 2025

  16. [16]

    Maleki, H

    A. Maleki, H. Radha. Faster-HEAL: An Efficient and Privacy- Preserving Collaborative Perception Framework for Heterogeneous Autonomous Vehicles. arXiv preprint arXiv:2603.07314. 2026 Mar 7

    arXiv 2026

  17. [17]

    Maleki, H

    A. Maleki, H. Radha. PEARL: A Lightweight Prompt-based Feature Interpreter Framework for Real-Time, Anonymous, and Heterogeneous Collaborative Perception. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition 2026 (pp. 1138-1147)

    2026

  18. [18]

    Perez, F

    E. Perez, F. Strub, H. De Vries, V . Dumoulin, A. Courville. FiLM: Visual reasoning with a general conditioning layer. In Proceedings of the AAAI conference on artificial intelligence 2018 Apr 29 (V ol. 32)

    2018

  19. [19]

    De Vries et al., Modulating early visual processing by language

    H. De Vries et al., Modulating early visual processing by language. Advances in neural information processing systems. 2017;30

    2017

  20. [20]

    J. Hu, L. Shen, G. Sun. Squeeze-and-excitation networks. In Pro- ceedings of the IEEE conference on computer vision and pattern recognition 2018 (pp. 7132-7141)

    2018

  21. [21]

    Jin et al., Lor-VP: Low-rank visual prompting for efficient vision model adaptation

    C. Jin et al., Lor-VP: Low-rank visual prompting for efficient vision model adaptation. arXiv preprint arXiv:2502.00896. 2025 Feb 2

    Pith/arXiv arXiv 2025

  22. [22]

    A. Li, L. Zhuang, S. Fan, S. Wang. Learning common and specific visual prompts for domain generalization. In Proceedings of the Asian conference on computer vision 2022 (pp. 4260-4275)

    2022

  23. [23]

    H. R. Medeiros, A. Belal, S. Muralidharan, E. Granger, M. Pedersoli. Visual modality prompt for adapting vision-language object detectors. In Proceedings of the IEEE/CVF International Conference on Com- puter Vision 2025 (pp. 2172-2182)

    2025

  24. [24]

    H. Phan, L. Tran, Q. Tran, T. Le. Enhancing domain adaptation through prompt gradient alignment. Advances in Neural Information Processing Systems. 2024 Dec 16;37:45518-51

    2024

  25. [25]

    S. Xie, R. Girshick, P. Doll´ar, Z. Tu, and K. He. Aggregated residual transformations for deep neural networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, 2017

    2017

  26. [26]

    Eckart- Young

    J. D. Carroll, J. J. Chang. Analysis of individual differences in multidimensional scaling via an N-way generalization of “Eckart- Young” decomposition. Psychometrika. 1970 Sep;35(3):283-319

    1970

  27. [27]

    Xu et al., V2x-vit: Vehicle-to-everything cooperative perception with vision transformer

    R. Xu et al., V2x-vit: Vehicle-to-everything cooperative perception with vision transformer. In European conference on computer vision, pages 107–124. Springer, 2022

    2022

  28. [28]

    Xu, et al., Opv2v: An open benchmark dataset and fusion pipeline for perception with vehicle-to-vehicle communication

    R. Xu, et al., Opv2v: An open benchmark dataset and fusion pipeline for perception with vehicle-to-vehicle communication. In 2022 In- ternational Conference on Robotics and Automation (ICRA), pp. 2583–2589. IEEE, 2022

    2022

  29. [29]

    A. H. Lang et al., Pointpillars: Fast encoders for object detection from point clouds. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 12697–12705, 2019

    2019

  30. [30]

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

    2018

  31. [31]

    Philion and S

    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 Aug 23 (pp. 194-210). Cham: Springer International Publishing

    2020

  32. [32]

    Tan and Q

    M. Tan and Q. Le. Efficientnet: Rethinking model scaling for con- volutional neural networks. In International conference on machine learning, pp. 6105–6114. PMLR, 2019

    2019

  33. [33]

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

    2016

  34. [34]

    Xu et al., Cobevt: Cooperative bird’s eye view semantic segmenta- tion with sparse transformers

    R. Xu et al., Cobevt: Cooperative bird’s eye view semantic segmenta- tion with sparse transformers. arXiv preprint arXiv:2207.02202, 2022

    Pith/arXiv arXiv 2022

  35. [35]

    L. V . Maaten, G. Hinton. Visualizing data using t-SNE. Journal of Machine Learning Research. 2008;9(Nov):2579-605

    2008

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.