Pith. sign in

REVIEW 4 major objections 7 minor 51 references

A Survey on Video Analytics in Cloud-Edge-Terminal Collaborative Systems

T0 review · 4 major / 7 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read This survey proposes the first structured taxonomy of video analytics in cloud-edge-terminal collaborative systems.

desk verdict Useful map of a fragmented field, but the 'systematic taxonomy' tag outruns the evidence until the categorization is checked. read the letter →

arxiv 2502.06581 v4 pith:GPVBD5MZ submitted 2025-02-10 cs.NI cs.CVcs.LG

Linxiao Gong , Hao Yang , Gaoyun Fang , Bobo Ju , Juncen Guo , Xiaoguang Zhu , Xiping Hu , Yan Wang
show 2 more authors
Peng Sun Azzedine Boukerche
This is my paper · ORCID
classification cs.NIcs.CVcs.LG
keywords videoanalyticscloud-edge-terminalcollaborationedgecomputingsurveytaxonomytaskoffloadingresource-awareschedulingmultimodallearning
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 is a survey that aims to organize the growing body of research on video analytics in cloud-edge-terminal collaborative (CETC) systems. It claims to be the first to present a structured taxonomy covering the full pipeline, from architectural paradigms (hierarchical, distributed, hybrid) through edge-centric and cloud-centric processing to hybrid task offloading and resource-aware scheduling. The authors argue that prior surveys focused on isolated pieces, such as edge computing or video streaming, and missed how the three tiers work together. If the taxonomy holds, it gives researchers and practitioners a shared map for positioning new work and spotting under-explored combinations of techniques, such as large language models with tiered offloading.

What carries the argument

The central organizing device is the taxonomy shown in Figure 2, which classifies existing works into edge-centric, cloud-centric, and hybrid video analytics, with subcategories such as on-device processing, edge-assisted offloading, model compression, training strategies, adaptive task offloading, and resource-aware scheduling. The taxonomy is built on a three-tier architecture model where terminal devices acquire video, edge nodes filter and preprocess, and the cloud trains models and runs complex inference; the paper also formalizes the tier interaction with simple cost and latency quantities such as total system latency $L_{sys} = L_{proc} + L_{trans} + L_{queue}$ and federated averaging $W^t = \frac{1}{N}\sum_{i=1}^N w_i^t$. The taxonomy's work is to give every surveyed paper a single slot, so that the field's coverage and gaps can be read off at a glance.

What would settle it

A systematic literature review with explicit search and inclusion criteria would settle the question: if it uncovers a substantial body of CETC video-analytics work that does not fit any taxonomy category, or finds that a random sample of the cited papers describe their own approach under a different framing than the category assigned here, then the taxonomy's completeness and the accuracy of its classifications fail.

Watch

Extended reading notes

Core claim

On its own terms, the paper's contribution is a taxonomy, not an empirical result. It divides video analytics in CETC systems into three top-level categories: edge-centric analytics (on-device processing, edge-assisted offloading, edge intelligence, edge-based video processing, and optimization strategies), cloud-centric analytics (cloud processing, training strategies, and inference optimization), and hybrid analytics (adaptive task offloading and resource-aware scheduling). Under each category it lists representative systems and techniques, and it identifies four open challenge areas: platform integration, system scalability, data protection, and resilience, with large language models and multimodal learning as emerging directions. The paper positions itself against earlier surveys that covered only edge computing, video streaming, or generative AI, and claims the first structured taxonomy that spans all three tiers.

Load-bearing premise

The survey assumes that the papers it selected are representative of the field and that each was accurately slotted into exactly one category of the taxonomy, without a systematic search protocol or inclusion criteria to back that selection.

Editorial extensions

If this is right

  • If the taxonomy is accepted, new CETC video analytics contributions can be positioned by category, making gaps such as cloud-centric edge preprocessing or hybrid model training visible.
  • The survey's challenge list gives a concrete agenda: platform integration, scalability, data protection, and resilience are named as open problems that block deployment.
  • Practitioners can use the taxonomy's representative systems (e.g., EdgeVision, ECIVA, Turbo, Shoggoth) as starting points when choosing between edge-centric and hybrid architectures.
  • The treatment of LLMs and multimodal integration as an emerging direction suggests that future CETC systems may pair foundation models with tiered offloading, an area the survey does not fully develop.

Reading between the lines

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

  • The survey's single-slot classification is an interpretation; a paper like Shoggoth, placed under edge-assisted offloading, also performs online learning, so a multi-label version of the taxonomy could reveal more overlap between categories.
  • The absence of a systematic search protocol means the taxonomy's balance may reflect the authors' network rather than the field; a bibliometric analysis of the cited venues could test this.
  • The taxonomy could be turned into a decision tool: given a deployment's latency, bandwidth, and privacy constraints, map the constraint vector to the category that publishes matching techniques, an extension the paper does not make.
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 / 7 minor

Summary. This manuscript surveys video analytics in cloud-edge-terminal collaborative (CETC) systems. It proposes a taxonomy (Fig. 2) that organizes prior work into edge-centric, cloud-centric, and hybrid analytics, and it reviews architectural paradigms, edge/cloud processing techniques, training and inference optimization, adaptive offloading and scheduling, followed by challenges and future directions around LLMs, scalability, privacy, and resilience. The paper's stated main contribution is being the first structured taxonomy spanning cloud, edge, and terminal tiers for video analytics.

Significance. If the taxonomy is accepted, the survey would provide a useful organizational frame for a fragmented literature and a convenient entry point; the breadth of recent references (2018–2024) is a genuine asset. The paper does not make empirical or mathematical claims, so the usual reproducibility criteria do not apply; its validity rests on accurate, representative, and complete description of the cited systems. The authors deserve credit for engaging with recent topics (LLM-based video understanding, federated learning, edge intelligence) and for illustrating the CETC workflow in Fig. 3. However, because the central contribution is a taxonomy, fidelity and methodological transparency of the classification are decisive; these are currently not established.

major comments (4)
  1. [Section 1, contribution 1; Figure 2] The claimed taxonomy 'across cloud, edge, and terminal tiers' is not realized in Fig. 2: the taxonomy has three top-level branches (edge-centric, cloud-centric, hybrid) and no terminal-centric branch; terminal devices appear only implicitly under edge-centric 'on-device processing.' This overstates what the survey actually organizes and should be either implemented (e.g., a terminal tier branch or an explicit mapping of papers to tiers) or softened in the contribution statement.
  2. [Figure 2, 'Adaptive Filtering' leaf; Section 3.5] EdgeDASH [Bayhan et al., 2021] is classified as edge-centric adaptive filtering, but the cited paper is about network-assisted adaptive video streaming and edge caching; Section 3.5 itself only says it 'resonates with' adaptive streaming. This is a misclassification by the paper's own description, and in the absence of any validation protocol it weakens confidence in the other taxonomy assignments.
  3. [Sections 1–6, methodology] The survey gives no search strategy, inclusion/exclusion criteria, database coverage, screening procedure, or coverage statistics. This is not merely a presentation gap: for a claim of 'first systematic taxonomy' and 'comprehensive review' (contribution 2), the absence of a reproducible selection protocol means a reader cannot assess completeness or bias. The authors should add a methodology subsection or at minimum a scope statement listing sources and selection rules.
  4. [Sections 3.4, 3.5, and similar one-sentence entries] Several survey entries are too thin to support the 'systematic' claim. For example, EdgeDASH is dispatched in a single sentence, and [Wu et al., 2020] and [Buniatyan, 2019] receive only parenthetical mentions. If the survey is to be a reference taxonomy, each included system needs at least a one-sentence statement of problem, mechanism, and reported result or limitation.
minor comments (7)
  1. [Section 6] There is a typo: 'In addtion' should be 'In addition.'
  2. [References, [Udrescu and Tegmark, 2021]] The reference title contains a misspelling: 'Symbolic pregression' should be 'Symbolic regression.'
  3. [Section 2.1] The inline optimization formulation 'min PN i=1 PM j=1 xij(cij + lij)' is not typeset and is difficult to read; a displayed equation with explicit constraints would improve clarity.
  4. [Section 3.4] The term 'OpenV AD' appears without definition or consistent formatting; it likely refers to open-set video anomaly detection and should be spelled out on first use.
  5. [Figure 1] The caption 'publication and citation statistics over the last decade' lacks axis labels, source database, and the query used to generate the counts.
  6. [Reference list] The reference list style is inconsistent: several entries (e.g., [Arnab et al., 2022], [Guo et al., 2023], [Jin et al., 2024]) omit venue/publisher information, while others include page ranges.
  7. [Section 4.3, Model Parallelism] CROSSBOW is introduced as a training-time technique under 'Inference Optimization'; this placement should be justified or moved to the training-related discussion.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the survey makes no empirical derivation claims; its taxonomy is a definitional organizational contribution, and the one overlapping-author citation is not load-bearing.

full rationale

This is a survey paper with no fitted parameters, no predictions, and no derivation chain from equations to conclusions. Its central contribution is the taxonomy in Figure 2, which is an organizational framework the authors impose on the literature; a taxonomy is definitional by nature, and the paper does not claim to derive the categories from properties of the cited systems. The only reference with overlapping authors is [Liu et al., 2024] (ACM CSUR), cited once in Sec. 4.2 as background for unsupervised anomaly detection; it supports a peripheral claim and is not load-bearing for the CETC taxonomy or any other central assertion. The placement of EdgeDASH under 'Adaptive Filtering' in Figure 2, while debatable, is a question of classification accuracy, not circularity: Sec. 3.5 only says EdgeDASH 'resonates with quality adaptation mechanisms in adaptive video streaming,' which is a descriptive characterization rather than a derivation. No passage asserts a limitation, missing support, or omitted proof that would alter the verdict. Accordingly, no circular step can be quoted or exhibited, and the appropriate score is 0.

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

The survey introduces no free parameters and no new entities. Its load-bearing inputs are the accuracy of the cited summaries and the representativeness of the selection. Both are plausible but unverified, and the paper provides no methodological evidence for them.

assumptions (3)
  • domain assumption The cited papers are accurately and fairly represented by the survey's one- or two-sentence summaries.
    Entire survey; e.g., Section 3.2 summarizes Turbo and ROI extraction in passing. If the summaries are inaccurate, the taxonomy misleads readers.
  • domain assumption The selection of references is representative of the field.
    No search protocol or inclusion criteria are given in Section 1 or in the methodology. The survey's completeness depends on this assumption.
  • domain assumption Taxonomy categories are mutually informative and not just arbitrary groupings.
    Figure 2's categorization (e.g., 'Edge-assisted Offloading' vs. 'Edge Intelligence') is asserted without criteria for assigning papers to categories.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Survey on Video Analytics in Cloud-Edge-Terminal Collaborative Systems." pith.science (2026). https://pith.science/paper/GPVBD5MZ

@misc{pith2026250206581,
  author       = {Pith},
  title        = {Pith review of: A Survey on Video Analytics in Cloud-Edge-Terminal Collaborative Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GPVBD5MZ}},
  note         = {Machine review of arXiv:2502.06581}
}
read the original abstract

The explosive growth of video data has driven the development of distributed video analytics in cloud-edge-terminal collaborative (CETC) systems, enabling efficient video processing, real-time inference, and privacy-preserving analysis. Among multiple advantages, CETC systems can distribute video processing tasks and enable adaptive analytics across cloud, edge, and terminal devices, leading to breakthroughs in video surveillance, autonomous driving, and smart cities. In this survey, we first analyze fundamental architectural components, including hierarchical, distributed, and hybrid frameworks, alongside edge computing platforms and resource management mechanisms. Building upon these foundations, edge-centric approaches emphasize on-device processing, edge-assisted offloading, and edge intelligence, while cloud-centric methods leverage powerful computational capabilities for complex video understanding and model training. Our investigation also covers hybrid video analytics incorporating adaptive task offloading and resource-aware scheduling techniques that optimize performance across the entire system. Beyond conventional approaches, recent advances in large language models and multimodal integration reveal both opportunities and challenges in platform scalability, data protection, and system reliability. Future directions also encompass explainable systems, efficient processing mechanisms, and advanced video analytics, offering valuable insights for researchers and practitioners in this dynamic field.

Figures

Figures reproduced from arXiv: 2502.06581 by the authors.

Figure 1
Figure 1. Publication and citation statistics over the last decade. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Taxonomy of video analytics in CETC systems. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Example of CETC video analytics workflow. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 48 canonical work pages

  1. [1]

    Be- yond transfer learning: Co-finetuning for action localisa- tion,

    [Arnab et al., 2022] Anurag Arnab, Xuehan Xiong, Alexey Gritsenko, Rob Romijnders, Josip Djolonga, Mostafa De- hghani, Chen Sun, Mario Luˇci´c, and Cordelia Schmid. Be- yond transfer learning: Co-finetuning for action localisa- tion,

    work page 2022

  2. [4]

    Hyper: Distributed cloud processing for large-scale deep learning tasks

    [Buniatyan, 2019] Davit Buniatyan. Hyper: Distributed cloud processing for large-scale deep learning tasks. In CSIT, pages 27–32,

    work page 2019

  3. [8]

    EdgeVision: Towards collaborative video ana- lytics on distributed edges for performance maximization

    [Gao et al., 2024] Guanyu Gao, Yuqi Dong, Ran Wang, and Xin Zhou. EdgeVision: Towards collaborative video ana- lytics on distributed edges for performance maximization. IEEE TMM, 26:9083–9094,

    work page 2024

  4. [10]

    Automatic pruning for quantized neural networks

    [Guerra and Drummond, 2021] Luis Guerra and Tom Drum- mond. Automatic pruning for quantized neural networks. In DICTA, pages 01–08,

    work page 2021

  5. [11]

    Fed-FSNet: Mitigating non-i.i.d

    [Guo et al., 2023] Jingcai Guo, Song Guo, Jie Zhang, and Ziming Liu. Fed-FSNet: Mitigating non-i.i.d. federated learning via fuzzy synthesizing network,

    work page 2023

  6. [12]

    Energy-efficient resource allocation for heterogeneous edge–cloud computing

    [Hua et al., 2024] Wei Hua, Peng Liu, and Linyu Huang. Energy-efficient resource allocation for heterogeneous edge–cloud computing. IEEE IoTJ, 11(2),

    work page 2024

  7. [13]

    Task offloading with enhanced deep q-networks for efficient industrial intelli- gent video analysis in edge–cloud collaboration

    [Ji et al., 2024] Xiaofeng Ji, Faming Gong, Nuanlai Wang, Chengze Du, and Xiangbing Yuan. Task offloading with enhanced deep q-networks for efficient industrial intelli- gent video analysis in edge–cloud collaboration. Adv. Eng. Inform., 62:102599,

    work page 2024

  8. [14]

    Video-lavit: Unified video-language pre-training with decoupled visual-motional tokenization,

    [Jin et al., 2024] Yang Jin, Zhicheng Sun, Kun Xu, Kun Xu, Liwei Chen, Hao Jiang, Quzhe Huang, Chengru Song, Yu- liang Liu, Di Zhang, Yang Song, Kun Gai, and Yadong Mu. Video-lavit: Unified video-language pre-training with decoupled visual-motional tokenization,

    work page 2024

Show all 51 references
  1. [15]

    Integration of edge–ai into iot–cloud architecture for landslide monitoring and prediction

    [Joshi et al., 2024] Amrita Joshi, Saurabh Agarwal, Debi Prasanna Kanungo, and Rajib Kumar Pani- grahi. Integration of edge–ai into iot–cloud architecture for landslide monitoring and prediction. IEEE TII , 20(3):4246–4258,

    2024

  2. [16]

    CROSSBOW: Scaling deep learning with small batch sizes on multi-gpu servers

    [Koliousis et al., 2019] Alexandros Koliousis, Pijika Watcharapichat, Matthias Weidlich, Luo Mai, Paolo Costa, and Peter Pietzuch. CROSSBOW: Scaling deep learning with small batch sizes on multi-gpu servers. PVLDB, 12(11):1399–1412,

    2019

  3. [17]

    Optimizing the energy con- sumption in three-tier cloud–edge–fog federated systems with omnidirectional offloading

    [Kuswiradyo et al., 2024] Primatar Kuswiradyo, Binayak Kar, and Shan-Hsiang Shen. Optimizing the energy con- sumption in three-tier cloud–edge–fog federated systems with omnidirectional offloading. CN, 250:110578,

    2024

  4. [18]

    Two- timescale joint service caching and resource allocation for task offloading with edge–cloud cooperation

    [Li et al., 2024] Yafei Li, Huiqiang Wang, Jiayu Sun, Hongwu Lv, Wenqi Zheng, and Guangsheng Feng. Two- timescale joint service caching and resource allocation for task offloading with edge–cloud cooperation. CN, 254:110771,

    2024

  5. [19]

    Efficient heterogeneous video seg- mentation at the edge

    [Lin et al., 2022] Jamie Menjay Lin, Siargey Pisarchyk, Juhyun Lee, David Tian, Tingbo Hou, Karthik Raveen- dran, Raman Sarokin, George Sung, Trent Tolley, and Matthias Grundmann. Efficient heterogeneous video seg- mentation at the edge. arXiv:2208.11666,

    2022 arXiv

  6. [20]

    Generalized video anomaly event detection: Sys- tematic taxonomy and comparison of deep models

    [Liu et al., 2024] Yang Liu, Dingkang Yang, Yan Wang, Jing Liu, Jun Liu, Azzedine Boukerche, Peng Sun, and Liang Song. Generalized video anomaly event detection: Sys- tematic taxonomy and comparison of deep models. ACM CSUR,

    2024

  7. [21]

    Unsupervised anomaly localization with structural feature-autoencoders

    [Meissen et al., 2023] Felix Meissen, Johannes Paetzold, Georgios Kaissis, and Daniel Rueckert. Unsupervised anomaly localization with structural feature-autoencoders. In Brainlesion, pages 14–24,

    2023

  8. [22]

    Revamp2t: Real-time edge video analytics for multicamera privacy-aware pedestrian tracking

    [Neff et al., 2020] Christopher Neff, Mat´ıas Mendieta, Shrey Mohan, Mohammadreza Baharani, Samuel Rogers, and Hamed Tabkhi. Revamp2t: Real-time edge video analytics for multicamera privacy-aware pedestrian tracking. IEEE IoTJ, 7(4):2591–2602,

    2020

  9. [24]

    Video-bench: A comprehensive benchmark and toolkit for evaluating video-based large language models,

    [Ning et al., 2023] Munan Ning, Bin Zhu, Yujia Xie, Bin Lin, Jiaxi Cui, Lu Yuan, Dongdong Chen, and Li Yuan. Video-bench: A comprehensive benchmark and toolkit for evaluating video-based large language models,

    2023

  10. [25]

    Ego-vehicle action recognition based on semi-supervised contrastive learning

    [Noguchi and Tanizawa, 2023] Chihiro Noguchi and Toshi- hiro Tanizawa. Ego-vehicle action recognition based on semi-supervised contrastive learning. In WACV, pages 5977–5987,

    2023

  11. [26]

    [Ollivier et al., 2023] S´ebastien Ollivier, Sheng Li, Yue Tang, Stephen Cahoon, Ryan Caginalp, Chayanika Chaud- huri, Peipei Zhou, Xulong Tang, Jingtong Hu, and Alex K. Jones. Sustainable ai processing at the edge. IEEE Micro, 43(1):19–28,

    2023

  12. [27]

    Video anomaly detection by estimating likelihood of representations

    [Ouyang and Sanchez, 2021] Yuqi Ouyang and Victor Sanchez. Video anomaly detection by estimating likelihood of representations. In ICPR,

    2021

  13. [28]

    Quan, Dinh C

    [Quan et al., 2023] Minh K. Quan, Dinh C. Nguyen, Van- Dinh Nguyen, Mayuri Wijayasundara, Sujeeva Setunge, and Pubudu N. Pathirana. HierSFL: Local differential privacy-aided split federated learning in mobile edge com- puting. In VCC, pages 103–108,

    2023

  14. [29]

    Perf4sight: A toolflow to model cnn training performance on edge gpus

    [Rajagopal and Bouganis, 2021] Aditya Rajagopal and Christos-Savvas Bouganis. Perf4sight: A toolflow to model cnn training performance on edge gpus. In ICCVW, pages 963–971,

    2021

  15. [30]

    Yadwadkar, and Christos Kozyrakis

    [Romero et al., 2021] Francisco Romero, Mark Zhao, Neer- aja J. Yadwadkar, and Christos Kozyrakis. Llama: A het- erogeneous & serverless framework for auto-tuning video analytics pipelines. In SoCC, pages 1–17,

    2021

  16. [31]

    Equilib- rium in the computing continuum through active inference

    [Sedlak et al., 2024] Boris Sedlak, Victor Casamayor Pujol, Praveen Kumar Donta, and Schahram Dustdar. Equilib- rium in the computing continuum through active inference. FGCS, 160:92–108,

    2024

  17. [32]

    [Sharma et al., 2024] Nelson Sharma, Aswini Ghosh, Rajiv Misra, and Sajal K. Das. Deep meta q-learning based multi-task offloading in edge-cloud systems. IEEE TMC, 23(4):2583–2598,

    2024

  18. [33]

    [Silva et al., 2021] Joaquim Silva, Eduardo R. B. Marques, Lu´ıs M.B. Lopes, and Fernando Silva. Energy-aware adap- tive offloading of soft real-time jobs in mobile edge clouds. J. Cloud Comput., 10(1):38,

    2021

  19. [34]

    Breckon, and Ioannis Katramados

    [Stephenson et al., 2019] Felix Stephenson, Toby P. Breckon, and Ioannis Katramados. Degraf-flow: Extending degraf features for accurate and efficient sparse-to-dense optical flow estimation. In ICIP, pages 1277–1281,

    2019

  20. [35]

    Dyannet: A scene dynamicity guided self-trained video anomaly detection network

    [Thakare et al., 2023] Kamalakar Vijay Thakare, Yash Raghuwanshi, Debi Prosad Dogra, Heeseung Choi, and Ig-Jae Kim. Dyannet: A scene dynamicity guided self-trained video anomaly detection network. In WACV, pages 5530–5539,

    2023

  21. [36]

    Dy- namic microservice deployment and offloading for things– edge–cloud computing

    [Tian et al., 2024] Xianzhong Tian, Huixiao Meng, Yifan Shen, Junxian Zhang, Yuzhe Chen, and Yanjun Li. Dy- namic microservice deployment and offloading for things– edge–cloud computing. IEEE IoTJ, 11(11):1–11,

    2024

  22. [37]

    Tran, Sattvic Ray, Linnea Lemma, Yunrui Li, Reef Sweeney, Aparna Baskaran, Zvonimir Dogic, Pengyu Hong, and Michael F

    [Tran et al., 2024] Phu N. Tran, Sattvic Ray, Linnea Lemma, Yunrui Li, Reef Sweeney, Aparna Baskaran, Zvonimir Dogic, Pengyu Hong, and Michael F. Hagan. Deep- learning optical flow outperforms piv in obtaining velocity fields from active nematics. arXiv:2404.15497,

    2024 arXiv

  23. [38]

    Symbolic pregression: Discovering physical laws from distorted video

    [Udrescu and Tegmark, 2021] Silviu-Marian Udrescu and Max Tegmark. Symbolic pregression: Discovering physical laws from distorted video. Phys. Rev. E , 103(4):043307,

    2021

  24. [39]

    Edge, fog, and cloud computing : An overview on challenges and applications,

    [V oet al., 2022] Thong V o, Pranjal Dave, Gaurav Bajpai, and Rasha Kashef. Edge, fog, and cloud computing : An overview on challenges and applications,

    2022

  25. [40]

    Shoggoth: Towards efficient edge-cloud collaborative real-time video inference via adaptive online learning

    [Wang et al., 2023] Liang Wang, Kai Lu, Nan Zhang, Xi- aoyang Qu, Jianzong Wang, Jiguang Wan, Guokuan Li, and Jing Xiao. Shoggoth: Towards efficient edge-cloud collaborative real-time video inference via adaptive online learning. In DAC, pages 1–6,

    2023

  26. [41]

    Descriptive and predictive analysis of aggregating functions in server- less clouds: The case of video streaming

    [Wu et al., 2020] Shangrui Wu, Chavit Denninnart, Xiangbo Li, Yang Wang, and Mohsen Amini Salehi. Descriptive and predictive analysis of aggregating functions in server- less clouds: The case of video streaming. In HPCC,

    2020

  27. [42]

    Transcoding-enabled cloud–edge–terminal collaborative video caching in heterogeneous iot networks: An online learning approach with time-varying information

    [Xiao et al., 2024] Han Xiao, Yirong Zhuang, Changqiao Xu, Wendong Wang, Hongke Zhang, Renjie Ding, Tengfei Cao, Lujie Zhong, and Gabriel-Miro Muntean. Transcoding-enabled cloud–edge–terminal collaborative video caching in heterogeneous iot networks: An online learning approac...

    2024

  28. [43]

    [Xu et al., 2019] Zhe Xu, Biao Min, and Ray C.C. Cheung. A robust background initialization algorithm with super- pixel motion detection. Signal Process.: Image Commun. , 71:1–12,

    2019

  29. [44]

    ig- niter: Interference-aware gpu resource provisioning for predictable dnn inference in the cloud

    [Xu et al., 2023] Fei Xu, Jianian Xu, Jiabin Chen, Li Chen, Ruitao Shang, Zhi Zhou, and Fangming Liu. ig- niter: Interference-aware gpu resource provisioning for predictable dnn inference in the cloud. IEEE TPDS , 34(3):812–827,

    2023

  30. [45]

    [Xu et al., 2024] Minrui Xu, Hongyang Du, Dusit Niyato, Ji- awen Kang, Zehui Xiong, Shiwen Mao, Zhu Han, Abbas Jamalipour, Dong In Kim, Xuemin Shen, Victor C. M. Le- ung, and H. Vincent Poor. Unleashing the power of edge- cloud generative ai in mobile networks: A survey of aig...

    2024

  31. [46]

    Denoising-based turbo message passing for compressed video background subtraction

    [Xue et al., 2021] Zhipeng Xue, Xiaojun Yuan, and Yang Yang. Denoising-based turbo message passing for compressed video background subtraction. IEEE TIP , 30:2682–2696,

    2021

  32. [47]

    A serverless cloud-fog platform for dnn-based video analytics with incremental learning

    [Zhang et al., 2021] Huaizheng Zhang, Meng Shen, Yizheng Huang, Yonggang Wen, Yong Luo, Guanyu Gao, and Kyle Guan. A serverless cloud-fog platform for dnn-based video analytics with incremental learning. arXiv:2102.03012,

    2021 arXiv

  33. [48]

    A survey on generative ai and llm for video generation, understanding, and streaming,

    [Zhou et al., 2024] Pengyuan Zhou, Lin Wang, Zhi Liu, Yan- bin Hao, Pan Hui, Sasu Tarkoma, and Jussi Kangasharju. A survey on generative ai and llm for video generation, understanding, and streaming,

    2024

  34. [49]

    Towards open set video anomaly detection

    [Zhu et al., 2022] Yuansheng Zhu, Wentao Bao, and Qi Yu. Towards open set video anomaly detection. In ECCV, pages 395–412,

    2022

  35. [50]

    Variance- covariance regularization improves representation learn- ing

    [Zhu et al., 2024a] Jiachen Zhu, Katrina Evtimova, Yubei Chen, Ravid Shwartz-Ziv, and Yann LeCun. Variance- covariance regularization improves representation learn- ing. arXiv:2306.13292,

    arXiv

  36. [51]

    Edge intelligence- assisted animation design with large models: A survey

    [Zhu et al., 2024b] Jing Zhu, Chuanjiang Hu, Edris Khezri, and Mohd Mustafa Mohd Ghazali. Edge intelligence- assisted animation design with large models: A survey. J. Cloud Comput., 13(1):48, 2024

    2024

  37. [2018]

    End- to-end learning of motion representation for video under- standing

    [Fan et al., 2018] Lijie Fan, Wenbing Huang, Chuang Gan, Stefano Ermon, Boqing Gong, and Junzhou Huang. End- to-end learning of motion representation for video under- standing. In CVPR, pages 6016–6025,

    2018

  38. [2019]

    Towards compute-optimal transfer learning

    [Caccia et al., 2023] Massimo Caccia, Alexandre Galashov, Arthur Douillard, Amal Rannen-Triki, Dushyant Rao, Michela Paganini, Laurent Charlin, Marc’Aurelio Ran- zato, and Razvan Pascanu. Towards compute-optimal transfer learning. arXiv:2304.13164,

    2023 arXiv

  39. [2020]

    Real-time index authentication for event-oriented surveillance video query using blockchain

    [Nikouei et al., 2018] Seyed Yahya Nikouei, Ronghua Xu, Deeraj Nagothu, Yu Chen, Alexander Aved, and Erik Blasch. Real-time index authentication for event-oriented surveillance video query using blockchain. In ISC2,

    2018

  40. [2021]

    Ekya: Continuous learning of video analytics mod- els on edge compute servers

    [Bhardwaj et al., 2022] Romil Bhardwaj, Zhengxu Xia, Ganesh Ananthanarayanan, Junchen Jiang, Yuanchao Shu, Nikolaos Karianakis, Kevin Hsieh, Paramvir Bahl, and Ion Stoica. Ekya: Continuous learning of video analytics mod- els on edge compute servers. In NSDI,

    2022

  41. [2022]

    EdgeDASH: Exploiting network-assisted adaptive video streaming for edge caching

    [Bayhan et al., 2021] Suzan Bayhan, Setareh Maghsudi, and Anatolij Zubow. EdgeDASH: Exploiting network-assisted adaptive video streaming for edge caching. IEEE TNSM, 18(2):1732–1745,

    2021

  42. [2023]

    Massively parallel video networks

    [Carreira et al., 2018] Jo˜ao Carreira, Viorica P ˘atr˘aucean, Laurent Mazare, Andrew Zisserman, and Simon Osindero. Massively parallel video networks. In ECCV,

    2018

  43. [2024]

    AI-enhanced cloud-edge- terminal collaborative network: Survey, applications, and future directions

    [Gu et al., 2024] Huixian Gu, Liqiang Zhao, Zhu Han, Gan Zheng, and Shenghui Song. AI-enhanced cloud-edge- terminal collaborative network: Survey, applications, and future directions. IEEE COMST, 26(2):1322–1385,

    2024

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.