Pith. sign in

REVIEW 3 major objections 6 minor 40 references

RoundaboutHD: High-Resolution Real-World Urban Environment Benchmark for Multi-Camera Vehicle Tracking

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read RoundaboutHD is a real-world 4K, four-camera benchmark for multi-camera vehicle tracking, annotating 512 vehicle identities and providing detection, tracking, and re-identification baselines.

desk verdict Useful new 4K roundabout MCVT dataset, but the ReID benchmark's train/test identities overlap, making Table 13's near-perfect scores meaningless. read the letter →

arxiv 2507.08729 v2 pith:RZKWYKRQ submitted 2025-07-11 cs.CV

Yuqiang Lin , Sam Lockyer , Mingxuan Sui , Li Gan , Florian Stanek , Markus Zarbock , Wenbin Li , Adrian Evans
show 1 more author
Nic Zhang
This is my paper · ORCID
classification cs.CV
keywords multi-cameravehicletrackingre-identificationbenchmarkdatasetroundabouttraffic4Kvideoobjectdetectionsingle-camerasmartcity
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

RoundaboutHD is a new open-source benchmark for multi-camera vehicle tracking (MCVT), built from four non-overlapping 4K cameras that record a real roundabout for ten minutes each. The paper's central claim is that this closes a gap left by existing public resources: CityFlow is lower-resolution and short per camera, while synthetic alternatives carry a domain gap. The dataset contains 512 manually linked vehicle identities and 549,909 verified bounding boxes, plus subsets for detection, single-camera tracking, and image-based re-identification. If the dataset works as described, researchers gain a public testbed for smart-city tasks such as anomaly detection, traffic density estimation, and suspect-vehicle tracking.

What carries the argument

The load-bearing object is the RoundaboutHD capture-and-annotation setup: four synchronized, non-overlapping 4K cameras placed around one roundabout, with identities manually linked across views to form the ground truth. A semi-automatic pipeline produces that ground truth—YOLOv12x detections are manually corrected, ResNet-101 appearance features and SMILEtrack trajectories are refined into single-camera tracks, and cross-camera identity links are built by hand from visual attributes and spatiotemporal consistency. The dataset also ships camera-to-GPS mappings built with the cameratransform package, vehicle attributes (type, colour, make, model), and an evaluation framework implementing AP, IDF1/MOTA, mAP/rank-k, and the AICITY-format IDF1 metric. This machinery is what lets the paper claim both realism and standardizability: the same scene structure that creates hard cases is also what makes the benchmark reproducible.

What would settle it

Inspect the released ReID identity lists and compute the intersection between the 510 training identities and the 310 test identities; a non-empty intersection would mean the near-perfect fine-tuned scores in Table 13 do not measure generalization to unseen vehicles, and a disjoint re-split would settle the benchmark's validity.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is the dataset itself: RoundaboutHD provides 40 minutes of simultaneously recorded, fully annotated 4K/15 fps footage from four non-overlapping cameras around a single roundabout, with 512 unique vehicle identities and 549,909 bounding boxes. The authors report that this gives the public a real-world MCVT resource with longer per-camera footage and higher resolution than CityFlow, and they support it with task-specific subsets: 549,909 detection boxes, 1,042 single-camera trajectories, and a 65,528-image ReID subset. Baseline experiments describe the dataset as demanding: per-camera detection mAP runs from about 39% to 78%, the best single-camera tracker averages 83.33 IDF1, fine-tuned ReID models reach near-perfect scores on the subset, and the ELECTRICITY multi-camera tracker reaches 28.14 IDF1, lower than on CityFlow or Synthehicle. The low MCVT score is presented as evidence that the roundabout's nonlinear trajectories and frequent occlusions are genuinely harder for current methods.

Load-bearing premise

The load-bearing premise is that the 510 training identities and the 310 test identities in the ReID subset are disjoint; the paper's reported total of 512 unique identities in the full dataset makes that premise impossible on the reported numbers.

Editorial extensions

If this is right

  • The dataset gives MCVT researchers a public yardstick with 4K resolution and ten-minute-per-camera footage, so methods can be tested on temporal consistency rather than on brief clips.
  • The multi-camera subset's low ELECTRICITY baseline of 28.14 IDF1 defines clear headroom for methods that handle nonlinear movement and occlusion inside a roundabout.
  • The ReID subset, with 65,528 images drawn from 510 training and 310 test identities, supports image-level vehicle retrieval studies alongside the video-level MCVT task.
  • The included vehicle attributes and camera-to-GPS mappings allow future work to combine appearance matching with geometry and vehicle-model priors without collecting new data.

Reading between the lines

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

  • A natural extension the paper leaves implicit is a day/night or weather variant: the current capture is one scene and likely one session, so the benchmark's difficulty profile is tied to a single set of lighting and traffic conditions.
  • Because the paper highlights a U-turn case where a vehicle exits one camera's view and later reappears in the same view, methods that assume monotonic travel between cameras should be tested explicitly for their handling of revisits on this dataset.
  • The camera-to-GPS mapping opens a direct test of whether learned spatiotemporal transition models can replace hand-annotated zones, which would make MCVT pipelines more portable across cities.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper introduces RoundaboutHD, a multi-camera vehicle tracking benchmark built from 40 minutes of 4K/15fps footage from four non-overlapping cameras around a US roundabout, with 549,909 annotated bounding boxes for 512 unique vehicle identities. It provides task subsets for detection, single-camera tracking, image-based ReID, and MCVT, together with vehicle attributes, camera calibration, a public evaluation framework, and baseline results using YOLOv12x/YOLOv11x, BOXMOT trackers, FastReID models, and ELECTRICITY.

Significance. RoundaboutHD is potentially a valuable public resource: it offers higher resolution and longer per-camera duration than CityFlow, includes rich contextual annotations, and ships evaluation code and baselines. The dataset statistics and detection/SCT/MCVT baseline numbers are largely internally consistent, and the release of the data and code is a concrete strength. However, the image-based ReID subset as specified has overlapping train and test identities, which invalidates the reported ReID generalization results. Because the ReID subset is advertised as a central component of the benchmark, this issue must be resolved before the paper's claims are accepted.

major comments (3)
  1. [§3.3, §3.1, Abstract] The reported ReID split is arithmetically inconsistent with the total identity count. Section 3.3 states that the ReID training set contains 41,128 images from 510 unique identities and the test set contains 24,400 images from 310 unique identities, while Section 3.1 and the Abstract state that the whole dataset contains 512 unique vehicle identities. If train and test identity sets were disjoint, their union would contain 820 identities, which is impossible within a 512-identity dataset. At least 308 identities must therefore appear in both sets. The paper offers no allowance for overlapping identities, so the advertised 'test' set does not evaluate generalization to unseen vehicles. The authors must construct a disjoint-identity split, document the overlap (or its absence), and state the protocol explicitly.
  2. [§4.3, Table 13] The near-perfect ReID results in Table 13 (mAP 97.86–99.19, Rank-1 97.76–99.66) are exactly what one would expect if test identities were present during training, since the model can memorize their appearances. These numbers therefore do not support the claim that the RoundaboutHD ReID subset is a valid benchmark. The ReID baselines must be rerun under a split with disjoint identities, and the new scores, along with the identity lists or overlap statistics, should be reported.
  3. [§4.3] The ReID training protocol selects the best checkpoint based on 'validation accuracy', but no validation split is described anywhere in Section 3.3 or Section 4.3. Please specify the validation identities (and ensure they are disjoint from the test identities) or replace this selection rule with a clearly defined protocol; otherwise the reported Table 13 results are not reproducible.
minor comments (6)
  1. [§1 vs §2.2] The Introduction says CityFlow is currently the only open-source dataset of this kind, while Section 2.2 says there are two publicly available vehicle-specified MTMCT datasets, CityFlow and Synthehicle. This contradiction should be corrected.
  2. [§3.3] The statement that each identity is represented by approximately 128.5 images appears to average across train and test images and across 512 total identities; please clarify the per-subset average or the denominator.
  3. [§3.3] The SCT ground truth is said to contain trajectories split by long-term occlusions leading to 'id switch'; please clarify how MOT metrics are computed when the ground truth itself contains multiple trajectories for one physical vehicle.
  4. [§4.4, Table 14] The ELECTRICITY distance threshold is fine-tuned on RoundaboutHD while the CityFlow and Synthehicle results are taken from other settings; please add a caveat that the cross-dataset comparison may be affected by this hyperparameter difference.
  5. [References] References [28] and [29] list the same paper; merge the duplicate citations.
  6. [Multiple sections] Typos: 'curial' (Section 4.1), 'serval' (Section 3.4), 'boarder' (Conclusion), 'associate's' (Introduction), and 'datasts' (Section 2) should be corrected.

Circularity Check

1 steps flagged · score 6.0 of 10

ReID subset's train and test identity sets must overlap by arithmetic, so Table 13's near-perfect scores are forced and do not measure generalization to unseen vehicles.

  1. fitted input called prediction [Section 3.1 (Dataset Overview) and Section 3.3 (Subset Overview), with results reported in Table 13]
    "Within the 40 minutes recording, a total of 549,909 bounding boxes of 512 vehicles are manually annotated. ... The training set includes 41,128 images from 510 unique identities. ... Among these, 23,227 gallery images and 1,173 query images from 310 unique identities form the testing set."

    Section 3.1 gives 512 total unique vehicle identities. Section 3.3 reports 510 unique identities in the ReID training set and 310 unique identities in the testing set. If these sets were disjoint, the combined count would be 820 identities, which exceeds the dataset-wide total of 512. The paper's own numbers therefore force at least 308 identities to appear in both training and testing. Models retrained on the RoundaboutHD training set in Table 13 are then evaluated on identities whose images were part of the training fit, so the near-perfect mAP (97.86–99.19) and Rank-1 (97.76–99.66) scores are inflated by construction. The advertised ReID benchmark does not measure generalization to unseen vehicles; the prediction shares its inputs with the fit.

full rationale

The paper is a dataset and benchmark paper rather than a derivation chain, so most claims are self-contained empirical statements. The central dataset description (40 minutes, four 4K cameras, 549,909 boxes, 512 identities) is not circular, and the detection, single-camera tracking, and MCVT baselines are normal self-evaluations on the proposed data. However, the image-based ReID subset has a concrete self-referential flaw: Section 3.3's identity counts (510 training, 310 test) contradict Section 3.1's total of 512 identities unless at least 308 identities occur in both train and test. Because the models in Table 13 were trained on those identities, the reported ReID 'predictions' share their inputs with the fit; the scores do not establish a valid re-identification benchmark. This is partial circularity affecting one advertised component, while the other benchmark components retain independent content. The tuning of ELECTRICITY's threshold on RoundaboutHD is a mild self-referential adjustment but does not make any claimed result equivalent to its input.

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

No new physical or mathematical entity is introduced; the ledger mainly records hand-chosen thresholds and domain assumptions in the data collection and evaluation pipeline.

free parameters (3)
  • ELECTRICITY cross-camera matching threshold = 12
    Fine-tuned on RoundaboutHD in Section 4.4; the reported MCVT baseline depends on this tuned value.
  • Annotation detection confidence threshold = 0.35
    Chosen in the semi-automatic pipeline in Section 3.2 to generate candidate boxes; affects which objects are initially labeled.
  • Minimum bounding box area for annotation = 900 px^2
    Set in Section 3.2 to suppress small/distant detections; changes the annotation census and downstream subset composition.
assumptions (3)
  • domain assumption Manual labeled annotations are treated as ground truth without a quantified label-error rate.
    Section 3.2 describes manual refinement but no inter-annotator agreement or error audit; every baseline is measured against this assumed ground truth.
  • domain assumption Four cameras are temporally synchronized.
    Section 3.1 states all videos begin recording simultaneously; drift over the 10-minute clips is not measured but is assumed negligible for cross-camera association.
  • standard math Standard MOT/ReID metrics (IDF1, MOTA, mAP, rank-k, mINP) are valid for this dataset.
    Metrics are adopted from prior challenges and used without derivation in Section 3.4.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RoundaboutHD: High-Resolution Real-World Urban Environment Benchmark for Multi-Camera Vehicle Tracking." pith.science (2026). https://pith.science/paper/RZKWYKRQ

@misc{pith2026250708729,
  author       = {Pith},
  title        = {Pith review of: RoundaboutHD: High-Resolution Real-World Urban Environment Benchmark for Multi-Camera Vehicle Tracking},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RZKWYKRQ}},
  note         = {Machine review of arXiv:2507.08729}
}
read the original abstract

The multi-camera vehicle tracking (MCVT) framework holds significant potential for smart city applications, including anomaly detection, traffic density estimation, and suspect vehicle tracking. However, current publicly available datasets exhibit limitations, such as overly simplistic scenarios, low-resolution footage, and insufficiently diverse conditions, creating a considerable gap between academic research and real-world scenario. To fill this gap, we introduce RoundaboutHD, a comprehensive, high-resolution multi-camera vehicle tracking benchmark dataset specifically designed to represent real-world roundabout scenarios. RoundaboutHD provides a total of 40 minutes of labelled video footage captured by four non-overlapping, high-resolution (4K resolution, 15 fps) cameras. In total, 512 unique vehicle identities are annotated across different camera views, offering rich cross-camera association data. RoundaboutHD offers temporal consistency video footage and enhanced challenges, including increased occlusions and nonlinear movement inside the roundabout. In addition to the full MCVT dataset, several subsets are also available for object detection, single camera tracking, and image-based vehicle re-identification (ReID) tasks. Vehicle model information and camera modelling/ geometry information are also included to support further analysis. We provide baseline results for vehicle detection, single-camera tracking, image-based vehicle re-identification, and multi-camera tracking. The dataset and the evaluation code are publicly available at: https://github.com/siri-rouser/RoundaboutHD.git

Figures

Figures reproduced from arXiv: 2507.08729 by the authors.

Figure 1
Figure 1. MCVT Problem Overview. The example shows two [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Illustration of the Mapping Process for Cam01. The [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. The Distribution of Vehicle Colors and Types in terms [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (2 more)
Figure 6
Figure 6. Figure 6: Example Challenging Cases for Object Detection in [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Rank-50 Retrieval Example for Image-Based Vehicle [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 29 canonical work pages

  1. [1]

    Bot- sort: Robust associations multi-pedestrian tracking

    Nir Aharon, Roy Orfaig, and Ben-Zion Bobrovsky. Bot- sort: Robust associations multi-pedestrian tracking. arXiv preprint arXiv:2206.14651, 2022. 6

    arXiv 2022

  2. [2]

    Multi-camera multi-object tracking: A review of current trends and future advances

    Temitope Ibrahim Amosa, Patrick Sebastian, Lila Iznita Izhar, Oladimeji Ibrahim, Lukman Shehu Ayinla, Abdulrah- man Abdullah Bahashwan, Abubakar Bala, and Yau Alhaji Samaila. Multi-camera multi-object tracking: A review of current trends and future advances. Neurocomputing, 552: 126558, 2023. 1

    work page 2023

  3. [3]

    Evaluating mul- tiple object tracking performance: the clear mot metrics

    Keni Bernardin and Rainer Stiefelhagen. Evaluating mul- tiple object tracking performance: the clear mot metrics. EURASIP Journal on Image and Video Processing, 2008:1– 10, 2008. 5

    2008

  4. [4]

    BoxMOT: Pluggable sota multi-object tracking modules for segmentation, object detection, and pose estimation

    Mikel Brostr ¨om. BoxMOT: Pluggable sota multi-object tracking modules for segmentation, object detection, and pose estimation. https : / / github . com / mikel - brostrom / boxmot, 2025. Commit version used: v13.0.17 (July 2025). 6

    work page 2025

  5. [5]

    Observation-centric sort: Rethink- ing sort for robust multi-object tracking

    Jinkun Cao, Jiangmiao Pang, Xinshuo Weng, Rawal Khi- rodkar, and Kris Kitani. Observation-centric sort: Rethink- ing sort for robust multi-object tracking. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9686–9696, 2023. 7

    work page 2023

  6. [6]

    Bench- marks for corruption invariant person re-identification.arXiv preprint arXiv:2111.00880, 2021

    Minghui Chen, Zhiqiang Wang, and Feng Zheng. Bench- marks for corruption invariant person re-identification.arXiv preprint arXiv:2111.00880, 2021. 5

    arXiv 2021

  7. [7]

    Mot20: A benchmark for multi object tracking in crowded scenes

    Patrick Dendorfer, Hamid Rezatofighi, Anton Milan, Javen Shi, Daniel Cremers, Ian Reid, Stefan Roth, Konrad Schindler, and Laura Leal-Taix ´e. Mot20: A benchmark for multi object tracking in crowded scenes. arXiv preprint arXiv:2003.09003, 2020. 2, 5

    arXiv 2003

  8. [8]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. 7

    work page 2009

Show all 40 references
  1. [9]

    CARLA: An open urban driving simulator

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. CARLA: An open urban driving simulator. In Proceedings of the 1st Annual Conference on Robot Learning, pages 1–16, 2017. 3

    2017

  2. [10]

    Cameratransform: a scientific python package for perspective camera corrections

    Richard Gerum, Sebastian Richter, Alexander Winterl, Ben Fabry, and Daniel Zitterbart. Cameratransform: a scientific python package for perspective camera corrections. arXiv preprint arXiv:1712.07438, 2017. 4

    2017 arXiv

  3. [11]

    Fast r-cnn

    Ross Girshick. Fast r-cnn. In Proceedings of the IEEE inter- national conference on computer vision , pages 1440–1448,

  4. [12]

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

    Haiyun Guo, Chaoyang Zhao, Zhiwei Liu, Jinqiao Wang, and Hanqing Lu. Learning coarse-to-fine structured feature embedding for vehicle re-identification. In Proceedings of the AAAI Conference on Artificial Intelligence, 2018. 2

    2018

  5. [13]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 4

    2016

  6. [14]

    Fastreid: A pytorch toolbox for general instance re-identification

    Lingxiao He, Xingyu Liao, Wu Liu, Xinchen Liu, Peng Cheng, and Tao Mei. Fastreid: A pytorch toolbox for general instance re-identification. arXiv preprint arXiv:2006.02631,

    2006 arXiv

  7. [15]

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

    Fabian Herzog, Junpeng Chen, Torben Teepe, Johannes Gilg, Stefan H ¨ormann, and Gerhard Rigoll. Synthehicle: Multi- vehicle multi-camera tracking in virtual cities. In Proceed- ings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 1–11, 2023. 2, 3

    2023

  8. [16]

    Vehicle re- identification in context

    Aytac ¸ Kanacı, Xiatian Zhu, and Shaogang Gong. Vehicle re- identification in context. In Pattern Recognition: 40th Ger- man Conference, GCPR 2018, Stuttgart, Germany, October 9-12, 2018, Proceedings 40, pages 377–390. Springer, 2019. 2

    2018

  9. [17]

    Yolov11: An overview of the key architectural enhancements

    Rahima Khanam and Muhammad Hussain. Yolov11: An overview of the key architectural enhancements. arXiv preprint arXiv:2410.17725, 2024. 6

    2024 arXiv

  10. [18]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer vision–ECCV 2014: 13th European conference, zurich, Switzerland, September 6-12, 2014, proceeding...

    2014

  11. [19]

    City-scale multi-camera vehicle tracking guided by cross- road zones

    Chong Liu, Yuqi Zhang, Hao Luo, Jiasheng Tang, Weihua Chen, Xianzhe Xu, Fan Wang, Hao Li, and Yi-Dong Shen. City-scale multi-camera vehicle tracking guided by cross- road zones. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 4129– ...

    2021

  12. [20]

    Deep relative distance learning: Tell the difference between similar vehicles

    Hongye Liu, Yonghong Tian, Yaowei Yang, Lu Pang, and Tiejun Huang. Deep relative distance learning: Tell the difference between similar vehicles. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 2167–2175, 2016. 2

    2016

  13. [21]

    Large- scale vehicle re-identification in urban surveillance videos

    Xinchen Liu, Wu Liu, Huadong Ma, and Huiyuan Fu. Large- scale vehicle re-identification in urban surveillance videos. In 2016 IEEE international conference on multimedia and expo (ICME), pages 1–6. IEEE, 2016. 2

    2016

  14. [22]

    A deep learning-based approach to progressive vehicle re- identification for urban surveillance

    Xinchen Liu, Wu Liu, Tao Mei, and Huadong Ma. A deep learning-based approach to progressive vehicle re- identification for urban surveillance. In Computer Vision– ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part II 14 , pa...

    2016

  15. [23]

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

    Yihang Lou, Yan Bai, Jun Liu, Shiqi Wang, and Lingyu 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, pages 3235–3243, 2019. 2

    2019

  16. [24]

    Bag of tricks and a strong baseline for deep person re-identification

    Hao Luo, Youzhi Gu, Xingyu Liao, Shenqi Lai, and Wei Jiang. Bag of tricks and a strong baseline for deep person re-identification. In Proceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition workshops, pages 0–0, 2019. 7

    2019

  17. [25]

    Deep oc-sort: Multi-pedestrian tracking by adaptive re-identification

    Gerard Maggiolino, Adnan Ahmad, Jinkun Cao, and Kris Kitani. Deep oc-sort: Multi-pedestrian tracking by adaptive re-identification. In 2023 IEEE International conference on image processing (ICIP), pages 3025–3029. IEEE, 2023. 6

    2023

  18. [26]

    Anastasiu, Zheng Tang, Ming-Ching Chang, Xiaodong Yang, Yue Yao, Liang 9 Zheng, Pranamesh Chakraborty, Christian E

    Milind Naphade, Shuo Wang, David C. Anastasiu, Zheng Tang, Ming-Ching Chang, Xiaodong Yang, Yue Yao, Liang 9 Zheng, Pranamesh Chakraborty, Christian E. Lopez, Anuj Sharma, Qi Feng, Vitaly Ablavsky, and Stan Sclaroff. The 5th ai city challenge. In The IEEE Conference on Compute...

    2021

  19. [27]

    Electricity: An efficient multi-camera vehicle track- ing system for intelligent city

    Yijun Qian, Lijun Yu, Wenhe Liu, and Alexander G Haupt- mann. Electricity: An efficient multi-camera vehicle track- ing system for intelligent city. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, pages 588–589, 2020. 8

    2020

  20. [28]

    Performance measures and a data set for multi-target, multi-camera tracking

    Ergys Ristani, Francesco Solera, Roger Zou, Rita Cucchiara, and Carlo Tomasi. Performance measures and a data set for multi-target, multi-camera tracking. In European conference on computer vision, pages 17–35. Springer, 2016. 3

    2016

  21. [29]

    Performance measures and a data set for multi-target, multi-camera tracking

    Ergys Ristani, Francesco Solera, Roger Zou, Rita Cucchiara, and Carlo Tomasi. Performance measures and a data set for multi-target, multi-camera tracking. In European conference on computer vision, pages 17–35. Springer, 2016. 5

    2016

  22. [30]

    Boost- track: boosting the similarity measure and detection confi- dence for improved multiple object tracking.Machine Vision and Applications, 35(3):53, 2024

    Vukasin D Stanojevic and Branimir T Todorovic. Boost- track: boosting the similarity measure and detection confi- dence for improved multiple object tracking.Machine Vision and Applications, 35(3):53, 2024. 6

    2024

  23. [31]

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

    Zheng Tang, Milind Naphade, Ming-Yu Liu, Xiaodong Yang, Stan Birchfield, Shuo Wang, Ratnesh Kumar, David Anastasiu, and Jenq-Neng Hwang. Cityflow: A city-scale benchmark for multi-target multi-camera vehicle tracking and re-identification. In Proceedings of the IEEE/CVF con- f...

    2019

  24. [32]

    Yolov12: Attention-centric real-time object detectors

    Yunjie Tian, Qixiang Ye, and David Doermann. Yolov12: Attention-centric real-time object detectors. arXiv preprint arXiv:2502.12524, 2025. 4, 5

    2025 arXiv

  25. [33]

    Smiletrack: Simi- larity learning for occlusion-aware multiple object tracking

    Yu-Hsiang Wang, Jun-Wei Hsieh, Ping-Yang Chen, Ming- Ching Chang, Hung-Hin So, and Xin Li. Smiletrack: Simi- larity learning for occlusion-aware multiple object tracking. In Proceedings of the AAAI Conference on Artificial Intelli- gence, pages 5740–5748, 2024. 4

    2024

  26. [34]

    Ua-detrac: A new benchmark and protocol for multi-object detection and tracking

    Longyin Wen, Dawei Du, Zhaowei Cai, Zhen Lei, Ming- Ching Chang, Honggang Qi, Jongwoo Lim, Ming-Hsuan Yang, and Siwei Lyu. Ua-detrac: A new benchmark and protocol for multi-object detection and tracking. Computer Vision and Image Understanding, 193:102907, 2020. 2

    2020

  27. [35]

    Traffic-informed multi-camera sensing (tims) system based on vehicle re-identification

    Hao Yang, Jiarui Cai, Meixin Zhu, Chenxi Liu, and Yinhai Wang. Traffic-informed multi-camera sensing (tims) system based on vehicle re-identification. IEEE transactions on in- telligent transportation systems, 23(10):17189–17200, 2022. 2, 3

    2022

  28. [36]

    Box-grained reranking matching for multi-camera multi-target tracking

    Xipeng Yang, Jin Ye, Jincheng Lu, Chenting Gong, Minyue Jiang, Xiangru Lin, Wei Zhang, Xiao Tan, Yingying Li, Xiaoqing Ye, et al. Box-grained reranking matching for multi-camera multi-target tracking. In Proceedings of the IEEE/CVF conference on computer vision and pattern rec...

    2022

  29. [37]

    Transformer for object re- identification: A survey

    Mang Ye, Shuoyi Chen, Chenyue Li, Wei-Shi Zheng, David Crandall, and Bo Du. Transformer for object re- identification: A survey. arXiv preprint arXiv:2401.06960,

    arXiv

  30. [38]

    Video-based multi-camera vehicle tracking via appearance-parsing spatio-temporal trajectory matching net- work

    Xiaoqin Zhang, Hongqi Yu, Yong Qin, Xiaolong Zhou, and Sixian Chan. Video-based multi-camera vehicle tracking via appearance-parsing spatio-temporal trajectory matching net- work. IEEE Transactions on Circuits and Systems for Video Technology, 2024. 2, 3

    2024

  31. [39]

    Bytetrack: Multi-object tracking by associating every detection box

    Yifu Zhang, Peize Sun, Yi Jiang, Dongdong Yu, Fucheng Weng, Zehuan Yuan, Ping Luo, Wenyu Liu, and Xinggang Wang. Bytetrack: Multi-object tracking by associating every detection box. In European conference on computer vision, pages 1–21. Springer, 2022. 6

    2022

  32. [40]

    Scalable person re-identification: A benchmark

    Liang Zheng, Liyue Shen, Lu Tian, Shengjin Wang, Jing- dong Wang, and Qi Tian. Scalable person re-identification: A benchmark. In Proceedings of the IEEE international con- ference on computer vision, pages 1116–1124, 2015. 3 10

    2015

Pith tools

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