Pith. sign in

REVIEW 4 major objections 6 minor 65 references

CLIPVehicle: A Unified Framework for Vision-based Vehicle Search

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims a single end-to-end network can jointly detect and identify vehicles in raw surveillance frames, replacing the separate detection-and-storage pipeline for vehicle search.

desk verdict A legitimate first benchmark for end-to-end vehicle search, but the SOTA claim rests on a new benchmark whose identity split is not verified and on comparisons that omit the obvious two-stage baseline. read the letter →

arxiv 2508.04120 v1 pith:ZT4DIU7C submitted 2025-08-06 cs.CV

Likai Wang , Ruize Han , Xiangqun Zhang , Wei Feng This is my paper · ORCID
classification cs.CV
keywords vehiclesearchjointdetectionandre-identificationvision-languagemodelsCLIPpromptlearningsurveillancevideobenchmark
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes CLIPVehicle, a single network that performs vehicle detection and vehicle re-identification together on raw surveillance frames, so that searching for a target vehicle no longer requires a separate stage that pre-detects and stores every vehicle patch. The paper argues that the conflict between detection, which needs shared vehicle commonness, and re-identification, which needs individual uniqueness, can be managed by a dual-granularity semantic-region alignment that borrows text-image alignment from CLIP, plus a multi-level identification loss at image, box, and feature levels. To test this, the authors build three vehicle-search benchmarks: one real-world dataset derived from CityFlowV2 and two synthetic datasets derived from Synthehicle. On all three, the paper reports that CLIPVehicle improves mAP and Top-1 accuracy over both vehicle Re-ID methods and person search methods adapted to the task.

What carries the argument

Two mechanisms carry the argument. The dual-granularity semantic-region alignment uses a frozen CLIP text encoder with two sets of prompts: object-granularity prompts ('a photo of a vehicle' / 'not a photo of a vehicle') supervise foreground-background discrimination, and ID-granularity prompts, formed as 'a photo of a [X]... vehicle with [X] color and [X] type' with learnable CoOp tokens, supervise identity classification through cosine-similarity losses $L_{sra} = L_{sra}^{obj} + L_{sra}^{id}$. The multi-level identification learning adds $L_{mil} = L_{mil}^{img} + L_{mil}^{box} + L_{mil}^{fea}$, a coarse-to-fine set of losses at whole-frame multi-label, ground-truth-box cross-entropy, and

What would settle it

Compute the overlap between the identity label sets of CityFlowV2 scenarios S01-S04 and S05. If any vehicle ID has bounding boxes on both sides of the split, the identity-disjointness assumption fails and the reported mAP/Top-1 numbers are not a clean measure of Re-ID generalization. A secondary check: rerun the Table II comparison after removing from the test set any identity that also appears in training, and see whether CLIPVehicle's advantage over COAT/OIMNet++ shrinks.

Watch

Extended reading notes

Core claim

The central claim is that vehicle search can be treated as an end-to-end detection-plus-identification problem rather than a sequential pipeline. The architecture follows a sequential ReID-by-detection design: a shared backbone produces region proposals, a detection head localizes vehicles, and an ID head extracts identity features from the detected boxes. The paper's specific contribution is to reconcile the two sub-tasks by aligning visual region features with text prompts at two granularities: object-granularity prompts ('a photo of a vehicle' versus not) reinforce detection, while ID-granularity prompts, with learnable tokens and attribute words for color and type, reinforce identity cla

Load-bearing premise

The benchmark's reported numbers rest on the assumption that no vehicle identity appears in both the training scenarios (S01-S04) and the test scenario (S05) of CityFlowVS; the paper states this disjointness but does not show that it was checked against vehicles that drive across the 16 intersections.

Editorial extensions

If this is right

  • Vehicle search can be run directly on raw video frames: the gallery is produced by the same network that performs matching, removing the separate detect-and-store pass.
  • Attribute knowledge (vehicle color, type) can be injected into identity features through text prompts, without requiring attribute labels at inference time.
  • The new CityFlowVS, SynVS-Day, and SynVS-All benchmarks give a common evaluation for end-to-end vehicle search, so future methods can be compared on the joint task rather than on detection and Re-ID separately.
  • Because the added losses are training-only, CLIPVehicle can be attached to existing sequential person-search baselines (e.g., SeqNet, COAT, OIMNet++) to raise their mAP without slowing inference.

Reading between the lines

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

  • Not in the paper: the same dual-granularity alignment could be applied to person search or generic object search, where the detection-vs-uniqueness conflict is the same; the paper's design choices are not vehicle-specific beyond the attribute prompts.
  • Not in the paper but directly testable: the CityFlowVS train/test identity-disjointness should be verified by checking CityFlowV2's per-vehicle camera and scenario traces; a single vehicle appearing in both S01-S04 and S05 would inflate the reported mAP.
  • Not in the paper: the synthetic datasets could serve as pre-training data for the real-world set, since the framework already uses a frozen Re-ID teacher; this cross-domain transfer is a natural next experiment the paper does not report.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes CLIPVehicle, a unified end-to-end framework for vision-based vehicle search that jointly performs vehicle detection and re-identification. The method combines a dual-granularity semantic-region alignment module using CLIP/CoOp text prompts (object-level foreground/background and ID-level class prompts) with a multi-level identification learning strategy (image-level multi-label classification, box-level classification, and feature-level distillation from a frozen Re-ID teacher). The authors also construct three vehicle search benchmarks: a real-world CityFlowVS and two synthetic datasets SynVS-Day and SynVS-All. Experiments compare against vehicle Re-ID and person search baselines and report consistent improvements in mAP and Top-1.

Significance. If validated, the work fills a practical gap: current vehicle search systems require a separate detection stage and patch storage, while CLIPVehicle aims to learn detection and Re-ID jointly. The proposed benchmark is a potentially valuable community resource, and the use of CLIP-style semantic alignment for joint detection and Re-ID is a reasonable research direction. The paper includes ablations, adaptability experiments, and a clear exposition of the training pipeline. However, the central superiority claim depends on benchmark validity, fair comparison protocols, and statistical robustness; these are not yet established.

major comments (4)
  1. [Sec. IV (CityFlowVS) / Sec. III-A] The claim that the benchmark measures generalization to unseen identities rests on Sec. III-A: 'the training set and test set do not have overlapping vehicle identities.' However, Sec. IV constructs the split by scenario (S01-S04 vs S05) and reports no check that the 329 training and 337 test identities are disjoint. CityFlowV2 is a multi-target multi-camera tracking dataset where vehicles are observed across cameras and intersections, so overlap is plausible. Please report identity-overlap statistics and, if overlap exists, re-split and re-run evaluations. This is load-bearing for all headline numbers on the real-world dataset.
  2. [Sec. V-A, Table II] Vehicle Re-ID baselines (CLIP-ReID, MSINet, MBR) are trained on ground-truth boxes but tested on boxes 'predicted by our method' (Sec. V-A). This is not a two-stage detection+Re-ID baseline: the detector is the proposed method's own detector, so the comparison does not establish that the unified framework outperforms a conventional pipeline with an independent detector. Please add a two-stage baseline (e.g., an independently trained detector plus a strong Re-ID model) and evaluate all Re-ID baselines on both GT boxes and a fixed independent detector's boxes.
  3. [Tables II-IV] The reported improvements over the strongest person-search baselines are small (e.g., CityFlowVS mAP 14.1 vs 13.3 for COAT; Top-1 83.8 vs 83.0 for OIMNet++; SynVS-All mAP 24.6 vs 23.7). No error bars, multiple seeds, or statistical tests are reported. Given the margins, it is unclear whether the main claim is robust. Please report mean/std over at least three runs or otherwise demonstrate significance.
  4. [Sec. III-C/E, Eqs. (5) and (8)] The ID-granularity alignment uses ID-specific text tokens learned on the training identities as class text embeddings; this is effectively a learned per-ID embedding and may not reflect semantic alignment from CLIP. The teacher Re-ID model used in Eq. (8) is trained on the same benchmark, and no teacher accuracy is reported. To support the claim that these components contribute, please report (i) teacher Re-ID performance, (ii) an ablation where the ID tokens are replaced by randomly initialized trainable embeddings without the CLIP text encoder, and (iii) an ablation without the distillation term. This would clarify whether gains come from CLIP's pretrained semantics or from extra parameters/training signals.
minor comments (6)
  1. [Throughout] There are typos such as 'vehcile' and 'vehciel' in Sec. III-D, and the abstract has awkward phrasing ('the researches on which using'). A language edit is recommended.
  2. [Eq. (4)] The IoU threshold for cdet_i is described as 'larger a threshold' without specifying the value. Please state the threshold explicitly.
  3. [Table I] The # Queries column is '-' for CityFlowVS, but Sec. IV says 1,759 queries were selected. Please fill in the number for consistency.
  4. [Sec. IV] The terms 'scenario' and 'camera' are used interchangeably in places. Clarify whether a scenario corresponds to a group of cameras and how the S01-S04/S05 split relates to camera coverage.
  5. [Table VI] FPS values are reported without specifying hardware details, batch size, or input resolution. Please provide a clear evaluation protocol for speed numbers.
  6. [Sec. III-D, Eq. (6)] The BCE loss expression has unbalanced parentheses in the printed text; verify the formula is correct.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the proposed training components are self-contained supervised losses, and the reported benchmark results are evaluated on a separate test split.

full rationale

The paper's core derivations are standard supervised learning objectives. The ID-specific text tokens are learned on training identities in a pre-training stage and then frozen; they are used as class embeddings in the ID-granularity alignment loss (Eq. 5) to train the identity head. This is analogous to using a classifier head, and no text token exists for test identities, so the final evaluation on test identities is not a re-fit of these learned tokens. The feature-level consistency distillation (Eq. 8) uses a frozen Re-ID teacher trained on the same training split; this is a standard distillation setup that does not inject test labels. The L1 loss pulls student features toward teacher features, but the teacher itself is derived from ground-truth training identities, not from the test set. There is no self-citation chain that carries the argument; the only author-overlapping citation ([15]) is in a related-work sentence about person Re-ID and is not load-bearing. The most plausible concern is that CityFlowVS is split by scenario (S01-S04 training, S05 testing) while the paper asserts disjoint vehicle identities, and this disjointness is not verified in the text. That is a benchmark validity or data-leakage risk, not a circular derivation: the asserted identity overlap is not used to construct any equation or to fit any parameter that is then called a prediction. Therefore, no step reduces to its own inputs by construction, and the circularity score is low.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The core supervised signals are all derived from the same training labels: the ID-specific text tokens are fitted to training identities, and the teacher Re-ID model is trained on the same data. There are no independent external benchmarks or pre-registered predictions, so the ledger is dominated by components fitted to the target distribution.

free parameters (3)
  • ID-specific prompt tokens [X]1..[X]M (M=4) = Learned from benchmark; not reported
    Randomly initialized, learned via contrastive loss in a pre-training stage on the training identities; used as class text embeddings in Eq. (5).
  • Teacher Re-ID model (ResNet50) for feature distillation = Trained on the same benchmark; weights not released
    Used as fixed target in Eq. (8); trained with standard vehicle Re-ID losses on the same data, making the distillation target self-derived.
  • Positive/negative IoU threshold for object-granularity alignment (Eq. 4) = Not specified, likely 0.5
    Region proposals are labeled as vehicle or background based on IoU with ground truth; the threshold is chosen implicitly.
assumptions (4)
  • domain assumption CLIP's pre-trained cross-modal embedding space remains discriminative for fine-grained vehicle identity classes when identity-specific text tokens are learned.
    The dual-granularity alignment (Sec. III-C) assumes the frozen text encoder plus learnable tokens can produce text embeddings that meaningfully separate vehicle identities.
  • ad hoc to paper A teacher Re-ID model trained on the same benchmark provides valid identification targets for the student.
    Feature-level distillation (Eq. (8)) uses a frozen ResNet50 teacher pre-trained on the same vehicle search training set as the target, not an external or stronger model.
  • domain assumption Training and test identities are disjoint.
    Sec. III-A states this property, but the benchmark construction (Sec. IV) splits by scenario without verifying that vehicles do not travel across train/test intersections.
  • domain assumption Vehicle detection and vehicle re-identification have fundamentally conflicting objectives that require a special unified design.
    This premise motivates the whole framework (Sec. I) but is never formally demonstrated or measured.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CLIPVehicle: A Unified Framework for Vision-based Vehicle Search." pith.science (2026). https://pith.science/paper/ZT4DIU7C

@misc{pith2026250804120,
  author       = {Pith},
  title        = {Pith review of: CLIPVehicle: A Unified Framework for Vision-based Vehicle Search},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZT4DIU7C}},
  note         = {Machine review of arXiv:2508.04120}
}
read the original abstract

Vehicles, as one of the most common and significant objects in the real world, the researches on which using computer vision technologies have made remarkable progress, such as vehicle detection, vehicle re-identification, etc. To search an interested vehicle from the surveillance videos, existing methods first pre-detect and store all vehicle patches, and then apply vehicle re-identification models, which is resource-intensive and not very practical. In this work, we aim to achieve the joint detection and re-identification for vehicle search. However, the conflicting objectives between detection that focuses on shared vehicle commonness and re-identification that focuses on individual vehicle uniqueness make it challenging for a model to learn in an end-to-end system. For this problem, we propose a new unified framework, namely CLIPVehicle, which contains a dual-granularity semantic-region alignment module to leverage the VLMs (Vision-Language Models) for vehicle discrimination modeling, and a multi-level vehicle identification learning strategy to learn the identity representation from global, instance and feature levels. We also construct a new benchmark, including a real-world dataset CityFlowVS, and two synthetic datasets SynVS-Day and SynVS-All, for vehicle search. Extensive experimental results demonstrate that our method outperforms the state-of-the-art methods of both vehicle Re-ID and person search tasks.

Figures

Figures reproduced from arXiv: 2508.04120 by the authors.

Figure 1
Figure 1. Illustration of vehicle detection, vehicle re-identification (Re-ID), [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the proposed end-to-end unified framework CLIPVehicle for vehicle search (training stage). Given a raw frame, we sequentially handle [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

65 extracted references · 62 canonical work pages

  1. [1]

    Multi-class vehicle detection using vdnet in heterogeneous traffic,

    P. Deshmukh, K. C. Rayasam, U. K. Sahoo, S. K. Das, and S. Majhi, “Multi-class vehicle detection using vdnet in heterogeneous traffic,” IEEE Trans. Intell. Transp. Syst. , vol. 25, no. 12, pp. 19 415–19 429, 2024

    work page 2024

  2. [2]

    A review of vehicle detection techniques for intelligent vehicles,

    Z. Wang, J. Zhan, C. Duan, X. Guan, P. Lu, and K. Yang, “A review of vehicle detection techniques for intelligent vehicles,” IEEE Trans. Neural Netw. Learn. Syst. , vol. 34, no. 8, pp. 3811–3831, 2022

    work page 2022

  3. [3]

    Scale region recognition network for object counting in intelligent transportation system,

    X. Guo, M. Gao, W. Zhai, Q. Li, and G. Jeon, “Scale region recognition network for object counting in intelligent transportation system,” IEEE Trans. Intell. Transp. Syst. , vol. 24, no. 12, pp. 15 920–15 929, 2023

    work page 2023

  4. [4]

    Nwpu-moc: A benchmark for fine-grained multi-category object counting in aerial images,

    J. Gao, L. Zhao, and X. Li, “Nwpu-moc: A benchmark for fine-grained multi-category object counting in aerial images,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1–14, 2024

    work page 2024

  5. [5]

    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,” inProceedings of the European Conference on Computer Vision , 2022, pp. 1–18

    work page 2022

  6. [6]

    Mutr3d: A multi- camera tracking framework via 3d-to-2d queries,

    T. Zhang, X. Chen, Y . Wang, Y . Wang, and H. Zhao, “Mutr3d: A multi- camera tracking framework via 3d-to-2d queries,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 4537–4546

    work page 2022

  7. [7]

    Multimodality adaptive transformer and mutual learning for unsupervised domain adaptation vehicle re-identification,

    X. Zhang, Y . Ling, K. Li, W. Shi, and Z. Zhou, “Multimodality adaptive transformer and mutual learning for unsupervised domain adaptation vehicle re-identification,” IEEE Trans. Intell. Transp. Syst. , 2024

    work page 2024

  8. [8]

    A Comprehensive Survey on Deep-Learning-based Vehicle Re-Identification: Models, Data Sets and Challenges

    A. Amiri, A. Kaya, and A. S. Keceli, “A comprehensive survey on deep-learning-based vehicle re-identification: Models, data sets and challenges,” arXiv preprint arXiv:2401.10643 , 2024

    work page Pith review arXiv 2024

Show all 65 references
  1. [9]

    Day-night cross-domain vehicle re-identification,

    H. Li, J. Chen, A. Zheng, Y . Wu, and Y . Luo, “Day-night cross-domain vehicle re-identification,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 12 626–12 635

    2024

  2. [10]

    Beyond human-level license plate super-resolution with progressive vehicle search and domain priori gan,

    W. Liu, X. Liu, H. Ma, and P. Cheng, “Beyond human-level license plate super-resolution with progressive vehicle search and domain priori gan,” in Proceedings of the ACM International Conference on Multimedia , 2017, pp. 1618–1626

    2017

  3. [11]

    Pvss: A progressive vehicle search system for video surveillance networks,

    X.-C. Liu, H.-D. Ma, and S.-Q. Li, “Pvss: A progressive vehicle search system for video surveillance networks,” Journal of Computer Science and Technology, vol. 34, pp. 634–644, 2019

    2019

  4. [12]

    Person search by a bi-directional task-consistent learning model,

    C. Wang, B. Ma, H. Chang, S. Shan, and X. Chen, “Person search by a bi-directional task-consistent learning model,” IEEE Trans. Multimedia, vol. 25, pp. 1190–1203, 2022

    2022

  5. [13]

    Person search challenges and solutions: A survey,

    X. Lin, P. Ren, Y . Xiao, X. Chang, and A. Hauptmann, “Person search challenges and solutions: A survey,” in Proceedings of the International Joint Conferences on Artificial Intelligence , 2021, pp. 4500–4507

    2021

  6. [14]

    Transg: transformer-based skeleton graph pro- totype contrastive learning with structure-trajectory prompted recon- struction for person re-identification,

    H. Rao and C. Miao, “Transg: transformer-based skeleton graph pro- totype contrastive learning with structure-trajectory prompted recon- struction for person re-identification,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 22...

    2023

  7. [15]

    A benchmark of video-based clothes-changing person re-identification,

    L. Wang, X. Zhang, R. Han, J. Yang, X. Li, W. Feng, and S. Wang, “A benchmark of video-based clothes-changing person re-identification,” arXiv preprint arXiv:2211.11165 , 2022

    2022 arXiv

  8. [16]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark et al., “Learning transferable visual models from natural language supervision,” in Proceedings of the International Conference on Machine Learning , 2021, pp. 8748–8763

    2021

  9. [17]

    Learning to prompt for vision- language models,

    K. Zhou, J. Yang, C. C. Loy, and Z. Liu, “Learning to prompt for vision- language models,” International Journal of Computer Vision , vol. 130, no. 9, pp. 2337–2348, 2022

    2022

  10. [18]

    Ensemblenet: A hybrid approach for vehicle detection and estimation of traffic density based on faster r-cnn and yolo models,

    U. Mittal, P. Chawla, and R. Tiwari, “Ensemblenet: A hybrid approach for vehicle detection and estimation of traffic density based on faster r-cnn and yolo models,” Neural Computing and Applications , vol. 35, no. 6, pp. 4755–4774, 2023

    2023

  11. [19]

    Robust and scalable vehicle re-identification via self-supervision,

    P. Khorramshahi, V . Shenoy, and R. Chellappa, “Robust and scalable vehicle re-identification via self-supervision,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 5295–5304

    2023

  12. [20]

    Influence of avc and hevc compression on detection of vehicles through faster r-cnn,

    P. H. Chan, A. Huggett, G. Souvalioti, P. Jennings, and V . Donzella, “Influence of avc and hevc compression on detection of vehicles through faster r-cnn,” IEEE Trans. Intell. Transp. Syst. , vol. 25, no. 1, pp. 203– 213, 2024

    2024

  13. [21]

    Smart traffic management of vehicles using faster r-cnn based deep learning method,

    A. Chaudhuri, “Smart traffic management of vehicles using faster r-cnn based deep learning method,” Scientific Reports, vol. 14, no. 1, p. 10357, 2024

    2024

  14. [22]

    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 Trans. Pattern Anal. Mach. Intell. , vol. 39, no. 6, pp. 1137–1149, 2016

    2016

  15. [23]

    Yolo-fa: Type-1 fuzzy attention based yolo detector for vehicle detection,

    L. Kang, Z. Lu, L. Meng, and Z. Gao, “Yolo-fa: Type-1 fuzzy attention based yolo detector for vehicle detection,” Expert Systems with Appli- cations, vol. 237, p. 121209, 2024

    2024

  16. [24]

    Lvd-yolo: An efficient lightweight vehicle detection model for intelligent transportation systems,

    H. Pan, S. Guan, and X. Zhao, “Lvd-yolo: An efficient lightweight vehicle detection model for intelligent transportation systems,” Image and Vision Computing , vol. 151, p. 105276, 2024

    2024

  17. [25]

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

    J. Redmon, S. Divvala, R. Girshick, and F. Ali, “You only look once: Unified, real-time object detection,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2016, pp. 779– 788. 9

    2016

  18. [26]

    Ssd: Single shot multibox detector,

    W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y . Fu, and A. C. Berg, “Ssd: Single shot multibox detector,” in Proceedings of the European Conference on Computer Vision , 2016, pp. 21–37

    2016

  19. [27]

    Large-scale vehicle re-identification in urban surveillance videos,

    X. Liu, W. Liu, H. Ma, and H. Fu, “Large-scale vehicle re-identification in urban surveillance videos,” in Proceedings of the IEEE International Conference on Multimedia & Expo , 2016, pp. 1–6

    2016

  20. [28]

    Veri-wild: A large dataset and a new method for vehicle re-identification in the wild,

    Y . Lou, Y . Bai, J. Liu, S. Wang, and L. Duan, “Veri-wild: A large dataset and a new method for vehicle re-identification in the wild,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 3235–3243

    2019

  21. [29]

    Learning coarse- to-fine structured feature embedding for vehicle re-identification,

    H. Guo, C. Zhao, Z. Liu, J. Wang, and H. Lu, “Learning coarse- to-fine structured feature embedding for vehicle re-identification,” in Proceedings of the AAAI Conference on Artificial Intelligence , 2018, pp. 6853–6860

    2018

  22. [30]

    Global-local discrimi- native representation learning network for viewpoint-aware vehicle re- identification in intelligent transportation,

    X. Chen, H. Yu, F. Zhao, Y . Hu, and Z. Li, “Global-local discrimi- native representation learning network for viewpoint-aware vehicle re- identification in intelligent transportation,” IEEE Trans. Instrum. Meas. , vol. 72, pp. 1–13, 2023

    2023

  23. [31]

    Hetero- geneous context interaction network for vehicle re-identification,

    K. Sun, X. Pang, M. Zheng, X. Nie, X. Li, H. Zhou, and Y . Yin, “Hetero- geneous context interaction network for vehicle re-identification,”Neural Networks, vol. 169, pp. 293–306, 2024

    2024

  24. [32]

    Relation preserving triplet mining for stabilising the triplet loss in re-identification systems,

    A. Ghosh, K. Shanmugalingam, and W.-Y . Lin, “Relation preserving triplet mining for stabilising the triplet loss in re-identification systems,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2023, pp. 4840–4849

    2023

  25. [33]

    Camera-aware differentiated clustering with focal contrastive learning for unsupervised vehicle re- identification,

    M. Qiu, Y . Lu, X. Li, and Q. Lu, “Camera-aware differentiated clustering with focal contrastive learning for unsupervised vehicle re- identification,” IEEE Trans. Circuits Syst. Video Technol., vol. 34, no. 10, pp. 10 121–10 134, 2024

    2024

  26. [34]

    A shortly and densely connected convolutional neural network for vehicle re- identification,

    J. Zhu, H. Zeng, Z. Lei, S. Liao, L. Zheng, and C. Cai, “A shortly and densely connected convolutional neural network for vehicle re- identification,” in Proceedings of the International Conference on Pat- tern Recognition, 2018, pp. 3285–3290

    2018

  27. [35]

    Vehiclenet: Learning robust visual representation for vehicle re-identification,

    Z. Zheng, T. Ruan, Y . Wei, Y . Yang, and T. Mei, “Vehiclenet: Learning robust visual representation for vehicle re-identification,” IEEE Trans. Multimedia, vol. 23, pp. 2683–2693, 2020

    2020

  28. [36]

    Vehicle re-identification based on keypoint segmentation of original image,

    Z. Hu, Y . Xu, R. S. P. Raj, X. Cheng, L. Sun, and L. Wu, “Vehicle re-identification based on keypoint segmentation of original image,” Applied Intelligence, vol. 53, no. 3, pp. 2576–2592, 2023

    2023

  29. [37]

    Git: Graph interactive transformer for vehicle re-identification,

    F. Shen, Y . Xie, J. Zhu, X. Zhu, and H. Zeng, “Git: Graph interactive transformer for vehicle re-identification,” IEEE Trans. Image Process. , vol. 32, pp. 1039–1051, 2023

    2023

  30. [38]

    Weakly supervised contrastive learning for unsupervised vehicle reidentification,

    J. Yu, H. Oh, M. Kim, and J. Kim, “Weakly supervised contrastive learning for unsupervised vehicle reidentification,” IEEE Trans. Neural Netw. Learn. Syst. , vol. 35, no. 11, pp. 15 543–15 553, 2024

    2024

  31. [39]

    A benchmark for vehicle re-identification in mixed visible and infrared domains,

    Q. Zhao, S. Zhan, R. Cheng, J. Zhu, and H. Zeng, “A benchmark for vehicle re-identification in mixed visible and infrared domains,” IEEE Signal Process. Lett. , vol. 31, pp. 726–730, 2024

    2024

  32. [40]

    Attribute-based vehicle search in crowded surveillance videos,

    R. Feris, B. Siddiquie, Y . Zhai, J. Petterson, L. Brown, and S. Pankanti, “Attribute-based vehicle search in crowded surveillance videos,” in Pro- ceedings of the ACM International Conference on Multimedia Retrieval , 2011, pp. 1–8

    2011

  33. [41]

    Sbnet: Segmentation-based network for natural language-based vehicle search,

    S. Lee, T. Woo, and S. H. Lee, “Sbnet: Segmentation-based network for natural language-based vehicle search,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops , 2021, pp. 4054–4060

    2021

  34. [42]

    Towards effective person search with deep learning: A survey from systematic perspective,

    P. Zhang, X. Yu, C. Wang, J. Zheng, X. Ning, and X. Bai, “Towards effective person search with deep learning: A survey from systematic perspective,” Pattern Recognition, vol. 152, p. 110434, 2024

    2024

  35. [43]

    Person search in a scene by jointly modeling people commonness and person uniqueness,

    Y . Xu, B. Ma, R. Huang, and L. Lin, “Person search in a scene by jointly modeling people commonness and person uniqueness,” in Proceedings of the ACM International Conference on Multimedia, 2014, pp. 937–940

    2014

  36. [44]

    Joint detection and identification feature learning for person search,

    T. Xiao, S. Li, B. Wang, L. Lin, and X. Wang, “Joint detection and identification feature learning for person search,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2017, pp. 3415–3424

    2017

  37. [45]

    Person re-identification in the wild,

    L. Zheng, H. Zhang, S. Sun, M. Chandraker, Y . Yang, and Q. Tian, “Person re-identification in the wild,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2017, pp. 1367–1376

    2017

  38. [46]

    Tcts: A task- consistent two-stage framework for person search,

    C. Wang, B. Ma, H. Chang, S. Shan, and X. Chen, “Tcts: A task- consistent two-stage framework for person search,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 11 952–11 961

    2020

  39. [47]

    Norm-aware embedding for efficient person search and tracking,

    D. Chen, S. Zhang, J. Yang, and B. Schiele, “Norm-aware embedding for efficient person search and tracking,” International Journal of Computer Vision, vol. 129, pp. 3154–3168, 2021

    2021

  40. [48]

    Bi-directional interaction network for person search,

    W. Dong, Z. Zhang, C. Song, and T. Tan, “Bi-directional interaction network for person search,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2020, pp. 2839–2848

    2020

  41. [49]

    Sequential end-to-end network for efficient person search,

    Z. Li and D. Miao, “Sequential end-to-end network for efficient person search,” in Proceedings of the AAAI Conference on Artificial Intelli- gence, vol. 35, no. 3, 2021, pp. 2011–2019

    2021

  42. [50]

    Joint discriminative representation learning for end-to-end person search,

    P. Zhang, X. Yu, X. Bai, C. Wang, J. Zheng, and X. Ning, “Joint discriminative representation learning for end-to-end person search,” Pattern Recognition, vol. 147, p. 110053, 2024

    2024

  43. [51]

    Cdul: Clip- driven unsupervised learning for multi-label image classification,

    R. Abdelfattah, Q. Guo, X. Li, X. Wang, and S. Wang, “Cdul: Clip- driven unsupervised learning for multi-label image classification,” in Proceedings of the International Conference on Computer Vision , 2023, pp. 1348–1357

    2023

  44. [52]

    Ov- track: Open-vocabulary multiple object tracking,

    S. Li, T. Fischer, L. Ke, H. Ding, M. Danelljan, and F. Yu, “Ov- track: Open-vocabulary multiple object tracking,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 5567–5577

    2023

  45. [53]

    Zegclip: Towards adapting clip for zero-shot semantic segmentation,

    Z. Zhou, Y . Lei, B. Zhang, L. Liu, and Y . Liu, “Zegclip: Towards adapting clip for zero-shot semantic segmentation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 11 175–11 185

    2023

  46. [54]

    Clip-reid: exploiting vision-language model for image re-identification without concrete text labels,

    S. Li, L. Sun, and Q. Li, “Clip-reid: exploiting vision-language model for image re-identification without concrete text labels,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 37, no. 1, 2023, pp. 1405–1413

    2023

  47. [55]

    Unveiling the power of clip in unsupervised visible-infrared person re-identification,

    Z. Chen, Z. Zhang, X. Tan, Y . Qu, and Y . Xie, “Unveiling the power of clip in unsupervised visible-infrared person re-identification,” in Proceedings of the ACM International Conference on Multimedia, 2023, pp. 3667–3675

    2023

  48. [56]

    Attribute and state guided structural embedding network for vehicle re-identification,

    H. Li, C. Li, A. Zheng, J. Tang, and B. Luo, “Attribute and state guided structural embedding network for vehicle re-identification,” IEEE Trans. Image Process., vol. 31, pp. 5949–5962, 2022

    2022

  49. [57]

    Aivr- net: Attribute-based invariant visual representation learning for vehicle re-identification,

    H. Zhang, Z. Kuang, L. Cheng, Y . Liu, X. Ding, and Y . Huang, “Aivr- net: Attribute-based invariant visual representation learning for vehicle re-identification,” Knowledge-Based Systems, vol. 289, p. 111455, 2024

    2024

  50. [58]

    Distilling the knowledge in a neural network,

    G. Hinton, “Distilling the knowledge in a neural network,” arXiv preprint arXiv:1503.02531, 2015

    2015 arXiv

  51. [59]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2016, pp. 770–778

    2016

  52. [60]

    Msinet: Twins contrastive search of multi-scale interaction for object reid,

    J. Gu, K. Wang, H. Luo, C. Chen, W. Jiang, Y . Fang, S. Zhang, Y . You, and J. Zhao, “Msinet: Twins contrastive search of multi-scale interaction for object reid,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 19 243–19 253

    2023

  53. [61]

    Strength in diversity: Multi-branch representation learning for vehicle re-identification,

    E. Almeida, B. Silva, and J. Batista, “Strength in diversity: Multi-branch representation learning for vehicle re-identification,” in Proceedings of the IEEE International Conference on Intelligent Transportation Systems, 2023, pp. 4690–4696

    2023

  54. [62]

    Cascade transformers for end-to-end person search,

    R. Yu, D. Du, R. LaLonde, D. Davila, C. Funk, A. Hoogs, and B. Clipp, “Cascade transformers for end-to-end person search,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 7267–7276

    2022

  55. [63]

    Oimnet++: Prototypical normalization and localization-aware learning for person search,

    S. Lee, Y . Oh, D. Baek, J. Lee, and B. Ham, “Oimnet++: Prototypical normalization and localization-aware learning for person search,” in Proceedings of the European Conference on Computer Vision , 2022, pp. 621–637

    2022

  56. [64]

    Cityflow: A city-scale benchmark for multi-target multi-camera vehicle tracking and re-identification,

    Z. Tang, M. Naphade, M.-Y . Liu, X. Yang, S. Birchfield, S. Wang, R. Ku- mar, D. Anastasiu, and J.-N. Hwang, “Cityflow: A city-scale benchmark for multi-target multi-camera vehicle tracking and re-identification,” in Proceedings of the IEEE/CVF Conference on Computer Vision an...

    2019

  57. [65]

    Synthehicle: Multi-vehicle multi-camera tracking in virtual cities,

    F. Herzog, J. Chen, T. Teepe, J. Gilg, S. H ¨ormann, and G. Rigoll, “Synthehicle: Multi-vehicle multi-camera tracking in virtual cities,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2023, pp. 1–11

    2023

Pith tools

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