Pith. sign in

REVIEW 4 major objections 4 minor 38 references

Real-time Video Target Tracking Algorithm Utilizing Convolutional Neural Networks (CNN)

T0 review · 4 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A CNN tracker with an optical-flow branch claims higher success in rapid-motion and occlusion scenes.

desk verdict The paper's central performance claim is entirely unsupported: the results section reports no numbers for the proposed algorithm, and two of the favorable figures are actually attributed to YOLOv3. read the letter →

arxiv 2411.18314 v1 pith:ECHJXRDI submitted 2024-11-27 cs.CV

Chaoyi Tan , Xiangtian Li , Xiaobo Wang , Zhen Qi , Ao Xiang This is my paper · ORCID
classification cs.CV
keywords convolutionalneuralnetworksreal-timevideotrackingopticalflowtargetdetectionandonlinelearninginter-frameregressionsurveillanceintelligenttransportation
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

This paper proposes a real-time video target tracking algorithm that couples a convolutional neural network detector with an optical-flow branch, an online model update mechanism, and a multi-task loss that predicts both single-frame boxes and inter-frame motion. The authors claim that this design tracks targets more successfully and fails less often than several mainstream tracking algorithms when targets move rapidly, are partially occluded, or sit in cluttered backgrounds. They report 25-30 FPS on GPU, which they argue is fast enough for live video surveillance and intelligent transportation. The paper's contribution is the specific integration of detection and tracking, not a new CNN architecture.

What carries the argument

The load-bearing mechanism is a joint detection-and-tracking network with two streams: a CNN feature branch for appearance and a Flow branch that processes optical flow between frames to supply temporal context. The Flow branch predicts the target's next-frame position and feeds an offset prediction module that refines the current detection. Training uses a multi-task loss under the R-FCN framework that combines a classification loss, a single-frame regression loss, and the added inter-frame regression loss for motion offsets; detection boxes follow YOLOv3's anchor-based scheme. Real-time speed comes from depthwise separable convolutions, bottleneck layers, cross-layer feature sharing, feature pyramids, and GPU-accelerated parallel frame processing.

What would settle it

Run the described algorithm and the claimed baselines (SIFT-based tracking, YOLOv3, YOLOv5, and a Transformer-based tracker) on a standardized video-tracking benchmark with annotated sequences covering rapid motion, partial occlusion, and background clutter, and measure success rate, recall, and FPS under the same protocol. If the plotted advantages cannot be reproduced, or the method does not beat the baselines on those metrics, the central claim is settled against the paper.

Watch

Extended reading notes

Core claim

The central claim is that combining CNN-based detection with inter-frame optical flow in a joint detection-tracking framework yields a tracker that is both accurate and real-time in challenging conditions. On the paper's account, the Flow branch encodes motion between consecutive frames and predicts the target's next position; an offset prediction module then refines the detected box using that motion feature map. The training objective extends the R-FCN multi-task loss with an inter-frame regression loss, and the whole pipeline is made fast with depthwise separable convolutions, bottleneck layers, feature pyramids, and GPU parallel processing. Experimental figures are said to show higher tracking success rates and recall than a SIFT-based tracker, and better time efficiency and loss convergence than YOLOv3 and a Transformer-based model, with an average of 25 FPS in dense multi-target scenes and 30 FPS on GPU.

Load-bearing premise

The load-bearing premise is that the figures in Section III are real experimental outputs from a fair comparison; the paper gives no dataset, no quantitative numbers, and no evaluation protocol, so if those curves do not reflect actual measurements, the central performance claim collapses.

Editorial extensions

If this is right

  • If the reported results hold, video surveillance systems can track targets through partial occlusion and rapid motion at frame rates suitable for live monitoring.
  • The inter-frame regression loss gives the tracker a temporal consistency cue that single-frame detectors lack, which is what the paper credits for higher recall in complex scenes.
  • The method reaches a practical real-time operating point (25-30 FPS on GPU) without the computational overhead the paper attributes to Transformer-based trackers.
  • The same joint detection-tracking design could be carried over to intelligent transportation tasks such as vehicle and pedestrian tracking, where the paper expects its main applications.

Reading between the lines

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

  • The paper does not report ablations, so the most direct next test would be to remove the Flow branch, the inter-frame loss, or the online update one at a time to see which component actually drives the claimed improvement.
  • Because the experimental evidence is presented only as figures with no dataset or metric values, a reader cannot yet compare this method against published numbers; re-running the same design on a public tracking benchmark with standardized evaluation would settle the comparison.
  • The optical-flow prior could plausibly be extended to long-term tracking, where targets disappear for many frames and a motion model alone cannot reacquire them, but the paper does not address full occlusion.
  • If the speed claim transfers to lighter CNN backbones, the approach might run on embedded devices, though that is not demonstrated in the paper.
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 / 4 minor

Summary. The paper proposes a real-time video target tracking algorithm that combines CNN-based detection with optical flow correlation features and online model updating, targeting scenarios with rapid motion, partial occlusion, and complex backgrounds. Section II describes a joint detection-and-tracking architecture with classification, single-frame regression, and inter-frame regression losses, presented as an extension of the R-FCN and YOLO families. Section III claims experimental verification through Figures 2–5, reporting higher tracking success rates, lower failure rates, and real-time speeds (25–30 FPS on GPU) against SIFT, YOLOv3/YOLOv5, and Transformer baselines. The manuscript contains no quantitative results, no dataset description, no evaluation metrics, no baseline versions, and no code, so the central performance claim is not supported by any checkable evidence.

Significance. If the claimed results were substantiated, the contribution would be an incremental engineering integration of known components (a flow branch added to an R-FCN/YOLO-style detector with an extra inter-frame regression loss). No parameter-free derivations, machine-checked proofs, reproducible code, or falsifiable quantitative predictions are provided; the paper's only empirical content is a set of bare figures with textual assertions. At present the manuscript cannot be independently evaluated or reproduced, and the central claim of improved tracking success and lower failure rates is unverified.

major comments (4)
  1. [Section III, Figures 2–5] The central performance claim is made entirely on the basis of Figures 2–5, but the text provides no dataset names, no metric definitions, no numerical results, no error bars, no baseline versions, and no evaluation protocol. Worse, the prose in Section III explicitly attributes the favorable comparisons in Figures 4 and 5 to YOLOv3: “Figure 4's detailed comparison highlights YOLOv3's significant advantage” and “Figure 5's in-depth comparison shows YOLOv3's notable advantages.” This does not test the proposed CNN+Flow algorithm, and if the figures are instead meant to show the proposed algorithm, the text does not say so. The claimed advantage in tracking success rate and failure rate is therefore not established.
  2. [Section II, Equations (1)–(7)] Equations (2)–(6) are the standard R-CNN/R-FCN coordinate parameterization for bounding-box prediction, and Equation (7) is the generic YOLOv3 decoding formula for tx, ty, tw, th. No derivation is given that connects these formulas to the proposed joint detection-and-tracking architecture, the Flow branch, the offset prediction module, or the online updating mechanism. Figure 1 is never described layer by layer, and hyperparameters, the loss weights λ1 and λ2, training data, optimizer settings, and implementation details are absent. The method is therefore underspecified and not reproducible as written.
  3. [Section III, FPS discussion] The speed claim of 25–30 FPS on GPU and 15 FPS on CPU cannot be assessed because no hardware model, video resolution, code version, or measurement methodology is reported. The statement that “YOLOv5 excels in speed” is presented in passing, but no YOLOv5 comparison is shown in any figure or table, and no quantitative speed values are given for the proposed algorithm and the baselines. The absence of a baseline table or a per-sequence breakdown makes the speed and robustness claims untestable.
  4. [Section II versus Section III] There is an unresolved inconsistency in the role of YOLOv3. Section II introduces the framework as an extension of R-FCN with three parallel branches and an inter-frame loss, while Section III discusses YOLOv3 as the subject of comparisons and claims. If YOLOv3 is intended to be the detection backbone of the proposed algorithm, that identification is never stated; if it is a separate baseline, the text does not say how its results are relevant to evaluating the proposed architecture. Either way, the paper fails to attribute the experimental curves to the method being proposed.
minor comments (4)
  1. [Equation (1)] Equation (1) is garbled by typesetting: the input dimension appears as “321”, the summation bound is inconsistent with the notation “x1, x2, x3”, and the variables Wi and bi are introduced without a clear indexing convention.
  2. [Figures 2–5] The figure captions are uninformative: they provide no axis labels, no units, no legend descriptions, and no explanation of what each curve represents. It is impossible to determine from the manuscript what quantity is being plotted or which method each line corresponds to.
  3. [Section II.A, “correlation features”] The term “correlation features” is used repeatedly, but it is never formally defined or distinguished from standard CNN feature maps. It is unclear whether this refers to cross-correlation of frame features, optical flow, or a different mechanism.
  4. [References] Several references are cited in contexts that do not match the corresponding entries; for example, the citation of [1] for reinforcement-learning decision-making, [8] for CNN-LSTM weather forecasting, and [21] for multi-UAV navigation do not connect to the specific claims they are attached to. The reference list also contains many entries that are unrelated to the technical content of the paper.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the equations are standard CNN/R-CNN/YOLOv3 components, and the performance claims are asserted from unreported figures rather than derived from fitted parameters or self-cited results.

full rationale

The paper's derivation chain uses standard CNN/R-CNN/YOLOv3 machinery: the multitask loss in Eq. (2) combines classification loss, single-frame regression loss, and an inter-frame regression loss; Eqs. (3)-(6) are the classic R-CNN coordinate parameterization for predicted and ground-truth offsets; Eq. (7) is the standard YOLOv3 anchor-decoding formula. None of these equations defines the claimed tracking success rate or failure rate in terms of a fitted parameter, and no fitted value is later relabeled as a prediction. The experimental assertions in Section III appeal to figures whose quantitative content is absent, and Figures 4 and 5 are explicitly described as showing YOLOv3's advantages rather than the proposed CNN-plus-flow algorithm; this is an evidentiary or non-sequitur problem, not a circular reduction. The self-citations present in the reference list are background citations offering general inspiration and are not load-bearing: no uniqueness theorem, no fitted ansatz, and no prior result is imported to force the paper's conclusions. Consequently, no circular step can be exhibited with a specific quote and reduction, and the appropriate circularity score is 0. The absence of dataset details, quantitative metrics, and baseline specifications is a completeness/verifiability limitation, not evidence of circularity.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central claim depends almost entirely on unstated implementation details and an unverified assumption that experiments were run. The only explicit free parameters are the loss weights in Eq. (2), which are never assigned values.

free parameters (2)
  • Loss weights λ1, λ2
    Appear in Eq. (2) but their values are not specified; they control the trade-off between classification, regression, and inter-frame losses and affect training and final performance.
  • Network hyperparameters
    Details such as number of layers, channels, anchor scales, and training schedule are omitted; these choices determine the behavior of the described network.
assumptions (3)
  • standard math Standard convolution and R-CNN/YOLO coordinate parameterization formulas are correct and applicable.
    The paper relies on Eqs. (1) to (7) as known background.
  • domain assumption Combining optical flow features with CNN features improves tracking robustness under occlusion and fast motion.
    Stated in Section II.A without derivation or citation of comparative evidence.
  • ad hoc to paper The described network was actually implemented and the figures reflect real experimental measurements.
    The paper provides no code, data, or protocol, so this is taken as an unverified assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Real-time Video Target Tracking Algorithm Utilizing Convolutional Neural Networks (CNN)." pith.science (2026). https://pith.science/paper/ECHJXRDI

@misc{pith2026241118314,
  author       = {Pith},
  title        = {Pith review of: Real-time Video Target Tracking Algorithm Utilizing Convolutional Neural Networks (CNN)},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ECHJXRDI}},
  note         = {Machine review of arXiv:2411.18314}
}
read the original abstract

Thispaperaimstoresearchandimplementa real-timevideotargettrackingalgorithmbasedon ConvolutionalNeuralNetworks(CNN),enhancingthe accuracyandrobustnessoftargettrackingincomplex scenarios.Addressingthelimitationsoftraditionaltracking algorithmsinhandlingissuessuchastargetocclusion,morphologicalchanges,andbackgroundinterference,our approachintegratestargetdetectionandtrackingstrategies.It continuouslyupdatesthetargetmodelthroughanonline learningmechanismtoadapttochangesinthetarget's appearance.Experimentalresultsdemonstratethat,when dealingwithsituationsinvolvingrapidmotion,partial occlusion,andcomplexbackgrounds,theproposedalgorithm exhibitshighertrackingsuccessratesandlowerfailurerates comparedtoseveralmainstreamtrackingalgorithms.This studysuccessfullyappliesCNNtoreal-timevideotarget tracking,improvingtheaccuracyandstabilityofthetracking algorithmwhilemaintaininghighprocessingspeeds,thus meetingthedemandsofreal-timeapplications.Thisalgorithm isexpectedtoprovidenewsolutionsfortargettrackingtasksin videosurveillanceandintelligenttransportationdomains.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 25 canonical work pages

  1. [1]

    Research on Autonomous Driving Decision-making Strategies based Deep Reinforcement Learning[J]

    Wang Z, Yan H, Wei C, et al. Research on Autonomous Driving Decision-making Strategies based Deep Reinforcement Learning[J]. arXiv preprint arXiv:2408.03084, 2024

    arXiv 2024

  2. [2]

    Gao H, Wang H, Feng Z, et al. A novel texture extraction method for the sedimentary structures’ classification of petroleum imaging logging[C]//Pattern Recognition: 7th Chinese Conference, CCPR 2016, Chengdu, China, November 5-7, 2016, Proceedings, Part II 7. Springer Singapore, 2016: 161-172

    work page 2016

  3. [3]

    Research on Improved U-net Based Remote Sensing Image Segmentation Algorithm

    Yang Q, Wang Z, Liu S, et al. Research on Improved U-net Based Remote Sensing Image Segmentation Algorithm[J]. arXiv preprint arXiv:2408.12672, 2024

    work page Pith review arXiv 2024

  4. [4]

    Machine Learning-Based Research on the Adaptability of Adolescents to Online Education[J]

    Wang M, Liu S. Machine Learning-Based Research on the Adaptability of Adolescents to Online Education[J]. arXiv preprint arXiv:2408.16849, 2024

    arXiv 2024

  5. [5]

    A Mixed-Heuristic Quantum-Inspired Simplified Swarm Optimization Algorithm for scheduling of real- time tasks in the multiprocessor system[J]

    Su P C, Tan S Y, Liu Z, et al. A Mixed-Heuristic Quantum-Inspired Simplified Swarm Optimization Algorithm for scheduling of real- time tasks in the multiprocessor system[J]. Applied Soft Computing, 2022, 131: 109807

    work page 2022

  6. [6]

    Zhong Y, Liu Y, Gao E, et al. Deep learning solutions for pneumonia detection: performance comparison of custom and transfer learning models[C]//International Conference on Automation and Intelligent Technology (ICAIT 2024). SPIE, 2024, 13401: 95-100

    work page 2024

  7. [7]

    A neural matrix decomposition recommender system model based on the multimodal large language model[J]

    Xiang A, Huang B, Guo X, et al. A neural matrix decomposition recommender system model based on the multimodal large language model[J]. arXiv preprint arXiv:2407.08942, 2024

    arXiv 2024

  8. [8]

    Deep Learning for Weather Forecasting: A CNN-LSTM Hybrid Model for Predicting Historical Temperature Data[J]

    Gong Y, Zhang Y, Wang F, et al. Deep Learning for Weather Forecasting: A CNN-LSTM Hybrid Model for Predicting Historical Temperature Data[J]. arXiv preprint arXiv:2410.14963, 2024

    arXiv 2024

Show all 38 references
  1. [9]

    A Multimodal Fusion Network For Student Emotion Recognition Based on Transformer and Tensor Product[J]

    Xiang A, Qi Z, Wang H, et al. A Multimodal Fusion Network For Student Emotion Recognition Based on Transformer and Tensor Product[J]. arXiv preprint arXiv:2403.08511, 2024

    2024 arXiv

  2. [10]

    Comparative analysis of X-ray image classification of pneumonia based on deep learning algorithm algorithm[J]

    Ma D, Yang Y, Tian Q, et al. Comparative analysis of X-ray image classification of pneumonia based on deep learning algorithm algorithm[J]

  3. [11]

    Comprehensive Overview of Artificial Intelligence Applications in Modern Industries[J]

    Weng Y, Wu J, Kelly T, et al. Comprehensive Overview of Artificial Intelligence Applications in Modern Industries[J]. arXiv preprint arXiv:2409.13059, 2024

    2024 arXiv

  4. [12]

    Federated generalized category discovery[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Pu N, Li W, Ji X, et al. Federated generalized category discovery[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024: 28741-28750

    2024

  5. [13]

    Feng S, Song R, Yang S, et al. U-net Remote Sensing Image Segmentation Algorithm Based on Attention Mechanism Optimization[C]//2024 9th International Symposium on Computer and Information Processing Technology (ISCIPT). IEEE, 2024: 633- 636

    2024

  6. [14]

    Leveraging artificial intelligence to enhance data security and combat cyber attacks[J]

    Weng Y, Wu J. Leveraging artificial intelligence to enhance data security and combat cyber attacks[J]. Journal of Artificial Intelligence General science (JAIGS) ISSN: 3006-4023, 2024, 5(1): 392-399

    2024

  7. [15]

    Research on Move-to-Escape Enhanced Dung Beetle Optimization and Its Applications[J]

    Feng S, Wang J, Li Z, et al. Research on Move-to-Escape Enhanced Dung Beetle Optimization and Its Applications[J]. Biomimetics, 2024, 9(9): 517

    2024

  8. [16]

    Fine-grained imbalanced leukocyte classification with global-local attention transformer[J]

    Chen B, Qin F, Shao Y, et al. Fine-grained imbalanced leukocyte classification with global-local attention transformer[J]. Journal of King Saud University-Computer and Information Sciences, 2023, 35(8): 101661

    2023

  9. [17]

    Dynamic conceptional contrastive learning for generalized category discovery[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Pu N, Zhong Z, Sebe N. Dynamic conceptional contrastive learning for generalized category discovery[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023: 7579- 7588

    2023

  10. [18]

    Infrared image super-resolution via lightweight information split network (2024)[EB/OL]

    Liu S, Yan K, Qin F, et al. Infrared image super-resolution via lightweight information split network (2024)[EB/OL]

    2024

  11. [19]

    Dynamic Fraud Detection: Integrating Reinforcement Learning into Graph Neural Networks[J]

    Dong Y, Yao J, Wang J, et al. Dynamic Fraud Detection: Integrating Reinforcement Learning into Graph Neural Networks[J]. arXiv preprint arXiv:2409.09892, 2024

    2024

  12. [20]

    A memorizing and generalizing framework for lifelong person re-identification[J]

    Pu N, Zhong Z, Sebe N, et al. A memorizing and generalizing framework for lifelong person re-identification[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023

    2023

  13. [21]

    DRAL: Deep Reinforcement Adaptive Learning for Multi-UAVs Navigation in Unknown Indoor Environment[J]

    Mo K, Chu L, Zhang X, et al. DRAL: Deep Reinforcement Adaptive Learning for Multi-UAVs Navigation in Unknown Indoor Environment[J]. arXiv preprint arXiv:2409.03930, 2024

    2024 arXiv

  14. [22]

    Improved Unet model for brain tumor image segmentation based on ASPP-coordinate attention mechanism[J]

    Wang Z, Chen Y, Wang F, et al. Improved Unet model for brain tumor image segmentation based on ASPP-coordinate attention mechanism[J]. arXiv preprint arXiv:2409.08588, 2024

    2024 arXiv

  15. [23]

    Predicting Employee Turnover in the Financial Company: A Comparative Study of CatBoost and XGBoost Models[J].2024

    Yin Z, Hu B, Chen S. Predicting Employee Turnover in the Financial Company: A Comparative Study of CatBoost and XGBoost Models[J].2024

    2024

  16. [24]

    An Ensemble Approach to Stock Price Prediction Using Deep Learning and Time Series Models[J]

    Sui M, Zhang C, Zhou L, et al. An Ensemble Approach to Stock Price Prediction Using Deep Learning and Time Series Models[J]. 2024

    2024

  17. [25]

    Recurrent neural network from adder’s perspective: Carry-lookahead RNN[J]

    Jiang H, Qin F, Cao J, et al. Recurrent neural network from adder’s perspective: Carry-lookahead RNN[J]. Neural Networks, 2021, 144: 297-306

    2021

  18. [26]

    Adaptive receptive field U-shaped temporal convolutional network for vulgar action segmentation[J]

    Cao J, Xu R, Lin X, et al. Adaptive receptive field U-shaped temporal convolutional network for vulgar action segmentation[J]. Neural Computing and Applications, 2023, 35(13): 9593-9606

    2023

  19. [27]

    Artistic Neural Style Transfer Algorithms with Activation Smoothing[J]

    Li X, Cao H, Zhang Z, et al. Artistic Neural Style Transfer Algorithms with Activation Smoothing[J]. arXiv preprint arXiv:2411.08014, 2024

    2024 arXiv

  20. [28]

    Fortifying the global data fortress: a multidimensional examination of cyber security indexes and data protection measures across 193 nations[J]

    Weng Y, Wu J. Fortifying the global data fortress: a multidimensional examination of cyber security indexes and data protection measures across 193 nations[J]. International Journal of Frontiers in Engineering Technology, 2024, 6(2): 13-28

    2024

  21. [29]

    Lifelong person re-identification via adaptive knowledge accumulation[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Pu N, Chen W, Liu Y, et al. Lifelong person re-identification via adaptive knowledge accumulation[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2021: 7901- 7910

    2021

  22. [30]

    Real-Time pill identification for the visually impaired using deep learning[J]

    Dang B, Zhao W, Li Y, et al. Real-Time pill identification for the visually impaired using deep learning[J]. arXiv preprint arXiv:2405.05983, 2024

    2024 arXiv

  23. [31]

    YOLO-PPA based efficient traffic sign detection for cruise control in autonomous driving[J]

    Zhang J, Zhang W, Tan C, et al. YOLO-PPA based efficient traffic sign detection for cruise control in autonomous driving[J]. arXiv preprint arXiv:2409.03320, 2024

    2024 arXiv

  24. [32]

    Distance Recomputator and Topology Reconstructor for Graph Neural Networks[J]

    Liu D, Jiang M. Distance Recomputator and Topology Reconstructor for Graph Neural Networks[J]. arXiv preprint arXiv:2406.17281, 2024

    2024

  25. [33]

    Big data and machine learning in defence[J]

    Weng Y. Big data and machine learning in defence[J]. International Journal of Computer Science and Information Technology, 2024, 16(2): 25-35

    2024

  26. [34]

    Autonomous navigation of unmanned vehicle through deep reinforcement learning[J]

    Xu L, Liu J, Zhao H, et al. Autonomous navigation of unmanned vehicle through deep reinforcement learning[J]. arXiv preprint arXiv:2407.18962, 2024

    2024 arXiv

  27. [35]

    TD3 Based Collision Free Motion Planning for Robot Navigation[J]

    Liu H, Shen Y, Zhou C, et al. TD3 Based Collision Free Motion Planning for Robot Navigation[J]. arXiv preprint arXiv:2405.15460, 2024

    2024 arXiv

  28. [36]

    Meta reconciliation normalization for lifelong person re-identification[C]//Proceedings of the 30th ACM international conference on multimedia

    Pu N, Liu Y, Chen W, et al. Meta reconciliation normalization for lifelong person re-identification[C]//Proceedings of the 30th ACM international conference on multimedia. 2022: 541-549

    2022

  29. [37]

    Deep Learning Powered Estimate of The Extrinsic Parameters on Unmanned Surface Vehicles[J]

    Shen Y, Liu H, Zhou C, et al. Deep Learning Powered Estimate of The Extrinsic Parameters on Unmanned Surface Vehicles[J]. arXiv preprint arXiv:2406.04821, 2024

    2024 arXiv

  30. [38]

    Deep Learning-Based Snore Sound Analysis for the Detection of Night-time Breathing Disorders[J]

    Dang B, Ma D, Li S, et al. Deep Learning-Based Snore Sound Analysis for the Detection of Night-time Breathing Disorders[J]

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.