Chameleon: scalable adaptation of video analytics
The DOI matches its Crossref record; 24 of the 25 references it cites were found in a registry or an index.
Abstract
Applying deep convolutional neural networks (NN) to video data at scale poses a substantial systems challenge, as improving inference accuracy often requires a prohibitive cost in computational resources. While it is promising to balance resource and accuracy by selecting a suitable NN configuration (e.g., the resolution and frame rate of the input video), one must also address the significant dynamics of the NN configuration's impact on video analytics accuracy. We present Chameleon, a controller that dynamically picks the best configurations for existing NN-based video analytics pipelines. The key challenge in Chameleon is that in theory, adapting configurations frequently can reduce resource consumption with little degradation in accuracy, but searching a large space of configurations periodically incurs an overwhelming resource overhead that negates the gains of adaptation. The insight behind Chameleon is that the underlying characteristics (e.g., the velocity and sizes of objects) that affect the best configuration have enough temporal and spatial correlation to allow the search cost to be amortized over time and across multiple video feeds. For example, using the video feeds of five traffic cameras, we demonstrate that compared to a baseline that picks a single optimal configuration offline, Chameleon can achieve 20-50% higher accuracy with the same amount of resources, or achieve the same accuracy with only 30--50% of the resources (a 2-3X speedup).
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 1×
“, [17, 18]) which have led to a continuous stream of increasingly accurate models for object detection and classification.”
BackgroundHe, K., Gkioxari, G., Dollar, P., & Girshick, R. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.322 · Cited by 29,137
Girshick, R. (2015) · 2015 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2015.169 · Cited by 28,105
The pascal visual object classes (VOC) challenge
Everingham, M., Van Gool, L., Williams, C. K. I., Winn, J., & Zisserman, A. (2010) · International Journal of Computer Vision · doi:10.1007/s11263-009-0275-4 · Cited by 19,599
MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
Howard, A. G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., Andreetto, M., & Adam, H. (2017) · arXiv · doi:10.48550/arxiv.1704.04861 · Cited by 19,015
Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning
Szegedy, C., Ioffe, S., Vanhoucke, V., & Alemi, A. (2017) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v31i1.11231 · Cited by 4,636
Speed/Accuracy Trade-Offs for Modern Convolutional Object Detectors
Huang, J., Rathod, V., Sun, C., Zhu, M., Korattikara, A., Fathi, A., Fischer, I., Wojna, Z., Song, Y., Guadarrama, S., et al. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.351 · Cited by 2,743 · cited 1×
“The inference models are implemented in Tensorflow and are pretrained on standard image datasets [23], and the switching of video frame rate and resolution is done by FFmpeg [10].”
Method(1993) · Choice Reviews Online · doi:10.5860/choice.31-0986 · Cited by 1,540
Analysis of Thompson Sampling for the multi-armed bandit problem
Agrawal, S., & Goyal, N. (2011) · arXiv (Cornell University) · doi:10.48550/arxiv.1111.1797 · Cited by 819 · cited 1×
“al [19] use Thompson sampling [11] to optimize the layout of a multi-component web page; they use greedy hill climbing to select the next layout, similar to Chameleon, but Chameleon exploits more independence structure and monotonicity.”
MethodErnest: efficient performance prediction for large-scale advanced analytics
Venkataraman, S., Yang, Z., Franklin, M. J. M., Recht, B., & Stoica, I. G. (2016) · Networked Systems Design and Implementation · Cited by 559
Self-citation
Research works printed but not linked: 2
- FFmpeg. http://ffmpeg.org/ , 2000 --2018. FFmpeg. http://ffmpeg.org/, 2000--2018.
- W. E. Hoover and M. Rockville . Algorithms for confidence circles and ellipses . Citeseer , 1984 . W. E. Hoover and M. Rockville. Algorithms for confidence circles and ellipses. Citeseer, 1984.
Cited by
Items 201–300 of 575
SmartEye: An Open Source Framework for Real-Time Video Analytics with Edge-Cloud Collaboration
Wang, X., & Gao, G. (2021) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3474085.3478330 · Cited by 14
“VideoStorm [6] and Chameleon [2] adopt video frame sampling and resolution downsizing to save the compute cost of inferences on cheap DNN models.”
BackgroundMaxim: DRL-Based Cross-Camera Streaming Configuration for Real-Time Video Analytics
Zhang, R., Zhou, Y., Wang, F., & Wang, Z. (2022) · IEEE International Conference on Multimedia and Expo (ICME) · doi:10.1109/icme52920.2022.9859614 · Cited by 14
“Moreover, since profiling is expensive, the gap between each profiling has to be large enough (usually 4 seconds [2]) to balance the profiling cost and the video dynamics, which can only achieve coarse-grained adaptation, especially with highly dynamic videos.”
BackgroundResultInfluentialECCVideo: A Scalable Edge Cloud Collaborative Video Analysis System
Han, Q., Ren, X., Zhao, P. F., Wang, Y., Wang, L., Zhao, C., & Yang, X. (2022) · IEEE Intelligent Systems · doi:10.1109/mis.2022.3214614 · Cited by 14
Learning-Based Query Scheduling and Resource Allocation for Low-Latency Mobile-Edge Video Analytics
Lin, J., Yang, P., Wu, W. R., Zhang, N., Han, T., & Yu, L. (2023) · IEEE Internet of Things Journal · doi:10.1109/jiot.2023.3300696 · Cited by 14
“Reducing bandwidth resources consumption distinctly decreases the analytic latency, but guaranteeing high-quality videos is also significant [28].”
BackgroundEdge-Assisted Adaptive Configuration for Serverless-Based Video Analytics
Wang, Z., Zhang, S., Cheng, J. L., Wu, Z., Cao, Z. Z., & Cui, Y. (2023) · IEEE International Conference on Distributed Computing Systems (ICDCS) · doi:10.1109/icdcs57875.2023.00058 · Cited by 14
Usas: A Sustainable Continuous-Learning´ Framework for Edge Servers
Mishra, C. S., Sampson, J., Kandemir, M., Narayanan, V., & Das, C. R. (2024) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca57654.2024.00073 · Cited by 14
“Prior works [12], [38], [68] have designed hyperparameter micro-profilers.”
MethodComAI: Enabling Lightweight, Collaborative Intelligence by Retrofitting Vision DNNs
Jayarajah, K., Wanniarachchige, D., Abdelzaher, T., & Misra, A. (2022) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom48880.2022.9796769 · Cited by 14
Context-Aware Streaming Perception in Dynamic Environments
Sela, G.-E., Gog, I., Wong, J., Agrawal, K. K., Mo, X., Kalra, S., Schafhalter, P., Leong, E., Wang, X., Balaji, B., et al. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-19839-7_36 · Cited by 14
Self-citationSemantic Indexes for Machine Learning-based Queries over Unstructured DataPreprint
Kang, D. D., Guibas, J., Bailis, P. D., Hashimoto, T., & Zaharia, M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2009.04540 · Cited by 14
FHVAC: Feature-Level Hybrid Video Adaptive Configuration for Machine-Centric Live Streaming
Zhang, Y., Zhang, W., Du, H., Yan, C., Liu, L., & Zheng, Q. (2024) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2024.3372046 · Cited by 13
“Furthermore, recent experiments in [12] have observed that individual knobs have independent impact on accuracy.”
BackgroundMethodInfluentialNeRFHub: A Context-Aware NeRF Serving Framework for Mobile Immersive Applications
Chen, B., Yan, Z., Han, B., & Nahrstedt, K. (2024) · International Conference on Mobile Systems, Applications, and Services · doi:10.1145/3643832.3661879 · Cited by 13
OctopInf: Workload-Aware Inference Serving for Edge Video Analytics
Nguyen, T.-T., Liebe, L., Tau, N.-Q., Wu, Y., Cheng, J., & Lee, D. (2025) · IEEE International Conference on Pervasive Computing and Communications · doi:10.1109/percom64205.2025.00032 · Cited by 13
“To reduce overhead, approaches like VideoStorm [27], [28], [29], [30], [31] exploit the resource-quality tradeoff by adjusting parameters such as encoding quality, frame rate, and resolution.”
Enabling microservices management for Deep Learning applications across the Edge-Cloud Continuum
Houmani, Z., Balouek‐Thomert, D., Caron, E., & Parashar, M. (2021) · International Symposium on Computer Architecture and High Performance Computing (SBAC-PAD) · doi:10.1109/sbac-pad53543.2021.00025 · Cited by 13
“Online/offline configurations adaptation is currently a promising solution to address the issue of limited resources [9], [26], [28]–[30].”
BackgroundMethodEdgeBOL: A Bayesian Learning Approach for the Joint Orchestration of vRANs and Mobile Edge AI
Ayala‐Romero, J. A., García‐Saavedra, A., Costa‐Pérez, X., & Iosifidis, G. (2023) · IEEE/ACM Transactions on Networking · doi:10.1109/tnet.2023.3268981 · Cited by 13
“For example, [21] and [22] search greedily in real-time for the most resource-prudent configuration; [23] and [24] allocate computing resources and decide the image compression (or, video quality) and neural network model; Although some works like [9], [10], and [25] consider the energy in MVA…”
BackgroundAQuA: A New Image Quality Metric for Optimizing Video Analytics Systems
Paul, S., Drolia, U., Hu, Y., & Chakradhar, S. (2023) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3568423 · Cited by 13
“, Chameleon [36], AWStream [102], and VideoStorm [103], proposed parameter tuning of resolution, frame-rate, and model used for analytics for better resource-accuracy tradeoff, such parameter tuning is applied after frame acquisition/registration by an edge camera.”
BackgroundMethodApproxNet: Content and Contention Aware Video Analytics System for the Edge.Preprint
Xu, R., Koo, J., Kumar, R., Bai, P., Mitra, S., Maghanath, G., & Bagchi, S. (2019) · arXiv (Cornell University) · Cited by 13
Task-agnostic Indexes for Deep Learning-based Queries over Unstructured Data.Preprint
Kang, D. D., Guibas, J., Bailis, P. D., Hashimoto, T., & Zaharia, M. (2020) · arXiv (Cornell University) · Cited by 13
Novas: Tackling Online Dynamic Video Analytics With Service Adaptation at Mobile Edge Servers
Zhang, L., Zhu, H., Fei, W., Li, Y., Zhang, M., Cao, J., & Guo, M. (2024) · IEEE Transactions on Computers · doi:10.1109/tc.2024.3416675 · Cited by 12
“The mainstream of video preprocessing methods [4], [5], [6], [7], [8], [9], [10], relying on offline performance profiling for resolution and frame sampling rate adjustment, may lead to unsatisfied analysis accuracy or intolerable delays when dealing with fast-changing video contents.”
BackgroundCheng, S., Feng, F., Bi, T., & Jiang, T. (2024) · IEEE Transactions on Cognitive Communications and Networking · doi:10.1109/tccn.2024.3519365 · Cited by 12
“In the area of video configuration decision-making, Jiang et al. [15] proposed Chameleon, which dynamically searches for the optimal configuration of video analysis pipelines, which enables efficient and scalable adaptive video analysis.”
Edge-Assisted Video Transmission with Adaptive Key Frame Selection: A Hierarchical DRL Approach
Zhu, W.-J., Chen, R., Yi, C., & Cai, J. (2023) · Biennial Symposium on Communications · doi:10.1109/bsc57238.2023.10201830 · Cited by 12
“Besides, the accuracy of video analysis implement at edge servers should be considered [8] , which is defined as ait = ωait +(1−ω)φit , where ω and 1−ω are the weight factors”
MethodHou, B., & Zhang, J. (2021) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn52387.2021.9533581 · Cited by 12
“VideoStorm [16] and Chameleon [17] personal data pipeline knobs to identify cheap and accurate configurations, and VideoEdge [18] also considers placement plans on the cluster hierarchy.”
BackgroundDuan, Z., Yang, Z., Samoilenko, R., Oza, D. S., Jagadeesan, A., Sun, M., Ye, H., Xiong, Z., Zussman, G., & Kostić, Z. (2021) · IEEE International Conference on High Performance Computing and Communications; IEEE International Conference on Smart City; IEEE International Conference on Data Science and Systems (HPCC/SmartCity/DSS) · doi:10.1109/hpcc-dss-smartcity-dependsys53884.2021.00226 · Cited by 12
“Many video analytics systems choose their overall optimal configurations by conducting profiling at the beginning of a video, while Chameleon system [8] keeps up with the intrinsic dynamics in videos by periodically profiling to find an optimal resource-accuracy trade-off.”
BackgroundEdge Assisted Real-time Instance Segmentation on Mobile Devices
Zhang, J., Huang, X., Xu, J., Wu, Y., Ma, Q., Miao, X., Zhang, L., Chen, P., & Yang, Z. (2022) · IEEE International Conference on Distributed Computing Systems (ICDCS) · doi:10.1109/icdcs54860.2022.00058 · Cited by 12
“Chameleon [14] and VideoStorm [48] apply video configurations to achieve higher accuracy with the same”
BackgroundMethodProvably Efficient Resource Allocation for Edge Service Entities Using Hermes
Zhang, S., Liang, Y., Ge, J., Xiao, M., & Wu, J. (2020) · IEEE/ACM Transactions on Networking · doi:10.1109/tnet.2020.2989307 · Cited by 12
“Chameleon [31] utilizes temporal persistence of top-k configurations and spatial similarities to minimize the resource consumption for video analysis.”
Chen, H., Song, M., Zhao, J., Dai, Y., & Li, T. (2019) · Proceedings - International Symposium on Computer Architecture/Proceedings · doi:10.1145/3307650.3322260 · Cited by 11
“Recent years, neural networks (NN) are gaining lots of interests as they achieve state-of-the-art performance on many tasks [4], including image recognition [5], speech recognition [6], video understanding [7] and analytics [8][9], autonomous driving [10][11].”
BackgroundRetractedRanked Window Query Retrieval over Video Repositories
Chen, Y., Yu, X., & Koudas, N. (2022) · IEEE International Conference on Data Engineering (ICDE) · doi:10.1109/icde53745.2022.00253 · Cited by 11
Efficient Transmission and Reconstruction of Dependent Data Streams via Edge Sampling
Wolfrath, J., & Chandra, A. (2022) · IEEE International Conference on Cloud Engineering · doi:10.1109/ic2e55432.2022.00013 · Cited by 11
“Camera deployments often exhibit temporal/spatial correlation and must contend with limited WAN resources [48].”
BackgroundENTRO: Tackling the Encoding and Networking Trade-off in Offloaded Video Analytics
Kim, S., Bin, K., Yang, D., Ha, S., Chong, S., & Lee, K. (2023) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3581783.3613785 · Cited by 11
Wang, Y., Zhao, K., Zhang, X., & Chen, X. (2024) · IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS) · doi:10.1109/infocomwkshps61880.2024.10620909 · Cited by 11
“For a specific model inference service, different image configurations offer varying resource-accuracy tradeoff [16].”
BackgroundOmniSense: Towards Edge-Assisted Online Analytics for 360-Degree Videos
Zhang, M., Zhu, Y., Shen, L., Wang, F., & Liu, J. (2023) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom53939.2023.10229105 · Cited by 11
“A variety of systems [5], [6], [21], [22] have been designed to automatically analyze the content of regular videos with vision models.”
End-Edge Coordinated Joint Encoding and Neural Enhancement for Low-Light Video Analytics
He, Y., Yang, P., Qin, T., & Zhang, N. (2023) · Globecom · doi:10.1109/globecom54140.2023.10436757 · Cited by 11
“For instance, Jiang et al. [8] and Liu et al. [9] show that dynamic adjustment of encoding parameters can significantly reduce bandwidth consumption without reducing analytics accuracy, while Wang et al. [10] show that adaptive control of resolution and frame rate is also beneficial to improve…”
BackgroundAmadeus: Scalable, Privacy-Preserving Live Video AnalyticsPreprint
D'souza, S., Bahl, V., Ao, L., & Cox, L. P. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.05163 · Cited by 11
MadEye: Boosting Live Video Analytics Accuracy with Adaptive Camera ConfigurationsPreprint
Wong, M. H., Ramanujam, M., Balakrishnan, G., & Netravali, R. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2304.02101 · Cited by 11
Utility-Aware Load Shedding for Real-time Video Analytics at the EdgePreprint
Saurez, E., Gupta, H. K., Röger, H., Bhowmik, S., Ramachandran, U., & Rothermel, K. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2307.02409 · Cited by 11
Region-based Content Enhancement for Efficient Video Analytics at the EdgePreprint
Wang, W., Mi, L., Cen, S., Dai, H., Li, Y., Fu, X., & Liu, Y. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2407.16990 · Cited by 11
“Prior studies, no matter server-feedback accurate configuration setup [56,69,101,103] or fast autonomy by video source (camera) itself [45,65,89, 91], addressed this issue well, saving considerable bandwidth resources but causing accuracy drop.”
Masykuroh, K., Hendrawan, H., Mulyana, E., & Krishna, F. (2025) · IEEE Access · doi:10.1109/access.2025.3561063 · Cited by 10
EdgeLeague: Camera Network Configuration With Dynamic Edge Grouping for Industrial Surveillance
Tu, J., Chen, C., Xu, Q., & Guan, X. (2022) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2022.3205938 · Cited by 10
InFi: End-to-End Learning to Filter Input for Resource-Efficiency in Mobile-Centric Inference
Yuan, M., Zhang, L., He, F., Tong, X., Song, M.-H., Xu, Z., & Li, X. (2023) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2023.3275981 · Cited by 10
“Focusing on object detection in video streams, Chameleon [71] proposes to adaptively select a suitable pipeline configuration including the resolution and frame rate of videos, backbone neural networks for inference, etc. Elf [11] is designed for mobile video analytics where the input data is…”
MethodSpatial and Temporal Detection With Attention for Real-Time Video Analytics at Edges
Yan, Y., Zhang, S., Jin, Y., Cheng, F., Qian, Z., & Lu, S. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3361016 · Cited by 10
“Thus, a common type of approaches in previous work attempts to reduce computational complexity, such as frame filtering, image downsampling, or model pruning [26], [27], [28].”
BackgroundBPS: Batching, Pipelining, Surgeon of Continuous Deep Inference on Collaborative Edge Intelligence
Hou, X., Guan, Y., Choi, N., & Han, T. (2024) · IEEE Transactions on Cloud Computing · doi:10.1109/tcc.2024.3399616 · Cited by 10
“Unlike systems that leverage spatial similarities among nearby cameras [22], our focus is on a more generalized Machine Learning as a Service (MLaaS) framework [1], [2], [3].”
BackgroundContent-Aware Joint Knob Configuration and Resource Allocation for Edge Video Analytics
Bai, T., Hou, B., Wang, Z., Liu, D., & Nallanathan, A. (2025) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2025.3533596 · Cited by 10
“The state-of-the-art on the joint knob configuration and resource allocation for edge video analytics are mainly designed for a specific scene based on a fixed configuration-accuracy pro-file [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27].”
Montpetit, M., & Crespi, N. (2021) · Wiley eBooks · doi:10.1002/9781119765554.ch10 · Cited by 10
Owl: A Pre-and Post-processing Framework for Video Analytics in Low-light Surroundings
Zhang, R.-X., Li, C., Wu, C., Huang, T., & Sun, L. (2023) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom53939.2023.10229059 · Cited by 10
Concerto: Client-server Orchestration for Real-Time Video Analytics
Li, C., Zhang, R.-X., Huang, T., Jia, L., & Sun, L. (2023) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3581783.3611770 · Cited by 10
RAVAS: Interference-Aware Model Selection and Resource Allocation for Live Edge Video Analytics
Rahmanian, A. A., Ali-Eldin, A., Tesfatsion, S. K., Skubic, B., Gustafsson, H., Shenoy, P. J., & Elmroth, E. (2023) · ACM/IEEE Symposium on Edge Computing (SEC) · doi:10.1145/3583740.3628443 · Cited by 10
“Some recent works aim to choose the best configurations for the object detection model and frame input resolution as the main knobs that impact both resource usage level and prediction accuracy[6].”
BackgroundMethodInfluentialEvaluating Multimedia Protocols on 5G Edge for Mobile Augmented Reality
Cao, J., Su, X. J., Finley, B., Pauanne, A., Ammar, M., & Hui, P. (2021) · International Conference on Mobility, Sensing and Networking (MSN) · doi:10.1109/msn53354.2021.00042 · Cited by 12
OneAdapt: Fast Adaptation for Deep Learning Applications via Backpropagation
Du, K., Liu, Y., Hao, Y., Zhang, Q., Wang, H., Huang, Y., Ananthanarayanan, G., & Jiang, J. (2023) · ACM Symposium on Cloud Computing (SoCC) · doi:10.1145/3620678.3624653 · Cited by 9
Self-citationAero: Adaptive Query Processing of ML Queries
Kakkar, G. T., Cao, J., Sengupta, A., Arulraj, J. J. P., & Kim, H. (2025) · Proceedings of the ACM on Management of Data · doi:10.1145/3725408 · Cited by 9
Energy-Efficient Classification at the Wireless Edge with Reliability Guarantees
Merluzzi, M., Battiloro, C., Lorenzo, P. D., & Strinati, E. C. (2022) · IEEE International Conference on Communications workshops · doi:10.1109/iccworkshops53468.2022.9814659 · Cited by 9
“As pointed out in [1], [4], several aspects must be tackled when performing edge inference tasks in the most efficient way, such as model compression techniques to reduce the memory footprint [5], model selection strategies to adaptively select the inference model online [6], or application oriented optimization, for…”
Gemini: a Real-time Video Analytics System with Dual Computing Resource Control
Lu, R., Hu, C., Wang, D., & Zhang, J. (2022) · IEEE/ACM Symposium on Edge Computing (SEC) · doi:10.1109/sec54971.2022.00020 · Cited by 9
ROMA: Run-Time Object Detection To Maximize Real-Time Accuracy
Lee, J., Varghese, B., & Vandierendonck, H. (2023) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv56688.2023.00634 · Cited by 9
“Deploying a single object detector limits real-time accuracy on dynamically varying video contents and compute resources due to the fixed structure of the detector, which is the motivation of run-time techniques [13, 18, 17] and our paper.”
BackgroundMethodInfluentialCrowdsourcing Upon Learning: Energy-Aware Dispatch With Guarantee for Video Analytics
Chen, Y., Zhang, S., Jin, Y., Qian, Z., Xiao, M., Li, W., Liang, Y., & Lu, S. (2023) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2023.3269807 · Cited by 9
“Forthevideoanalyticswith Forthevideoanalyticswith manualannotationsorimagelabelling,theconfigurationsneed tobedeterminedcarefully,sincetheaccuracyofobjectdetec-tion[12]reliesonthem,includingframerates,resolutions,etc.”
BackgroundMethodInfluentialAn Edge-Side Real-Time Video Analytics System With Dual Computing Resource Control
Hu, C., Lu, R., Sang, Q., Liang, H., Wang, D., Cheng, D., Zhang, J., Li, Q., & Peng, J. (2023) · IEEE Transactions on Computers · doi:10.1109/tc.2023.3301136 · Cited by 9
“…AWS DeepLens camera with Intel Atom CPU and Intel Gen9 GPU The rocket system runs the built-in filtering algorithm [9], and the configu-ration algorithm [10] to select video resolutions from 144, 280, 540, 720, 1080p and the NN model from pre-trained models of TinyYOLO, SSD300, RetinaNet, and YOLOv3.”
BackgroundMethodOAVS: Efficient Online Learning of Streaming Policies for Drone-sourced Live Video Analytics
Li, Z., Zhang, M., & Zhu, Y. (2024) · International Workshop on Quality of Service · doi:10.1109/iwqos61813.2024.10682870 · Cited by 9
ChatCam: Embracing LLMs for Contextual Chatting-to-Camera with Interest-Oriented Video Summarization
Xiao, K., Gao, Y. C., Li, F., Xu, W., Chen, P., & Dong, W. (2024) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3699731 · Cited by 9
Adaptive Model Scheduling for Resource-efficient Data Labeling
Yuan, M., Zhang, L., Li, X., Yang, L., & Xiong, H. (2022) · ACM Transactions on Knowledge Discovery from Data · doi:10.1145/3494559 · Cited by 9
Towards a Methodology for Building Dynamic Urgent Applications on Continuum Computing Platforms
Balouek‐Thomert, D., Caron, E., Lefèvre, L., & Parashar, M. (2022) · doi:10.1109/ciw-ius56691.2022.00009 · Cited by 9
“Systems in [24], [27] adopt an online configuration adaptation algorithms for video analytics in Edge computing.”
WiseCam: Wisely Tuning Wireless Pan-Tilt Cameras for Cost-Effective Moving Object Tracking
Jinlong, E., He, L., Li, Z., & Liu, Y. (2023) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom53939.2023.10228926 · Cited by 9
Enabling high frame-rate UHD real-time communication with frame-skipping
Wang, T., Meng, Z., Xu, M., Han, R., & Liu, H. (2021) · doi:10.1145/3477083.3481582 · Cited by 9
Hybrid mobile vision for emerging applications
Wu, N., Lin, F. X., Qian, F., & Han, B. (2022) · IEEE Workshop on Mobile Computing Systems and Applications (WMCSA) · doi:10.1145/3508396.3512876 · Cited by 9
“, VideoStorm [47] and Chameleon [15]), inference accuracy (e.”
BackgroundTowards End-to-End Latency Guarantee in MEC Live Video Analytics with App-RAN Mutual Awareness
Yi, J., Lee, G., Jeong, M., Shin, S. H., Kim, D., & Lee, Y. (2025) · International Conference on Mobile Systems, Applications, and Services · doi:10.1145/3711875.3729139 · Cited by 8
SplitStream: Distributed and workload-adaptive video analytics at the edge
Liang, Y., Zhang, S., & Wu, J. (2024) · Journal of Network and Computer Applications · doi:10.1016/j.jnca.2024.103866 · Cited by 8
Mystique: User-Level Adaptation for Real-Time Video Analytics in Edge Networks via Meta-RL
Shi, X., Zhang, S., Liu, M., Meng, L., Liu, W., Gu, Y., Liu, K., Cheng, H., Song, Y., Lei, T., et al. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3514088 · Cited by 8
“QoE pattern diversity: Firstly, existing works [23], [24], [25], [26] typically integrate a fixed QoE pattern into the configuration adaptation algorithm for live video analytics.”
InfluentialZhao, M., Liu, S., Wu, F., & Chen, G. (2025) · ACM Conference on Embedded Networked Sensor Systems (SenSys) · doi:10.1145/3715014.3722044 · Cited by 8
Split Convolutional Neural Networks for Distributed Inference on Concurrent IoT Sensors
Chen, J., Le, D. V., Tan, R., & Ho, D. (2021) · IEEE International Conference on Parallel and Distributed Systems (ICPADS) · doi:10.1109/icpads53394.2021.00014 · Cited by 8
“Existing studies (e.g., [20]) have proposed various approaches to adapt the frame rate at runtime with the objective of avoiding capturing consecutive frames with the same contents.”
BackgroundMethodInfluentialSplit Computing Video Analytics Performance Enhancement With Auction-based Resource Management
Fu, K.-J., Yang, Y., & Wei, H. (2022) · IEEE Access · doi:10.1109/access.2022.3211984 · Cited by 8
“In video analytics systems, the accuracy profile, i.e., the relation between accuracy and resource consumption or con-figuration knobs, is video content varying [14], [15], [16].”
BackgroundResultInfluentialProfiling-free Configuration Adaptation and Latency-Aware Resource Scheduling for Video Analytics
Zhou, T., Wu, F., & Gao, L. (2022) · IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata55660.2022.10020573 · Cited by 8
“Different from these profiling-based adaptation methods [7], [8], [10]–[12], we propose a profiling-free adaptation that selects a configAbstract—With increasingly deployed cameras and the rapid advances of Computer Vision, large-scale live video analytics becomes feasible.”
BackgroundMethodQuickAdapt: Scalable Adaptation for Big Data Cyber Security Analytics
Ullah, F., & Babar, M. A. (2019) · IEEE International Conference on Engineering of Complex Computer Systems (ICECCS) · doi:10.1109/iceccs.2019.00016 · Cited by 8
“Several recent studies [25-27] also faced the same issue of large adaptation time and proposed solutions accordingly.”
BackgroundChen, N., Zhang, S., Quan, S., Ma, Z., Qian, Z., & Lu, S. (2021) · Computer Networks · doi:10.1016/j.comnet.2021.108513 · Cited by 8
Multitask Deep Learning for Edge Intelligence Video Surveillance System
Li, J., Zheng, Z., Li, Y., Ma, R., & Xia, S. (2020) · IEEE International Conference on Industrial Informatics (INDIN) · doi:10.1109/indin45582.2020.9442166 · Cited by 8
“The other approaches, such as the Chameleon [18] and VideoEdge [19] pick the best configuration parameters of edge devices for different deep learning algorithm pipelines.”
MethodBoggart: Towards General-Purpose Acceleration of Retrospective Video AnalyticsPreprint
Agarwal, N., & Netravali, R. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.15315 · Cited by 11
“…queries on live video, with optimizations along the following axes: (1) profiling pipeline knobs to identify cheaper (but accurate) configurations [59, 101], (2) integrating on-camera or edge server resources for partial inference or filtering [31, 35,41,74,92,103], (3) content/model-aware…”
BackgroundMachine Learning Systems for Highly-Distributed and Rapidly-Growing Data
Hsieh, K. (2019) · arXiv (Cornell University) · doi:10.1184/r1/10262450 · Cited by 8
RIVA: Communication-Efficient Streaming Control for Real-Time Industrial Video Analytics
Li, Z., Zhu, Y., Mumtaz, S., Kong, L., & Li, B. (2025) · IEEE Journal on Selected Areas in Communications · doi:10.1109/jsac.2025.3574591 · Cited by 7
“For example, Chameleon [44] uses dynamic configuration profiling to manage backend computational resources during video inference.”
Inferencing on Edge Devices: A Time- and Space-aware Co-scheduling Approach
Pereira, D., Ghose, A., Ghosh, S., & Dey, S. (2023) · ACM Transactions on Design Automation of Electronic Systems · doi:10.1145/3576197 · Cited by 7
“We argue that such an assumption is not viable for real-life scenarios where there exists an inherent time-varying demand for deploying NN pipelines with different configurations of weight tensors [22].”
BackgroundFair DNN Model Selection in Edge AI via A Cooperative Game Approach
Xie, J., Zhou, Z., Ouyang, T., Zhang, X., & Chen, X. (2023) · IEEE International Conference on Distributed Computing Systems (ICDCS) · doi:10.1109/icdcs57875.2023.00063 · Cited by 7
“However, when applying model selection to scale the edge AI system serving multiple applications at the edge, the existing work merely optimizes the trade-off between system-wide inference accuracy and system cost [8] [9] [10] [11].”
BackgroundMethodQoS-aware Resource Optimization for Hierarchical Cross-Edge Video Analytics
Zhao, K., Zhou, Z., Ouyang, T., Zhao, M., Wang, H., & Chen, X. (2023) · IEEE International Conference on Sensing, Communication, and Networking (SECON) · doi:10.1109/secon58729.2023.10287440 · Cited by 7
“Compared with the configuration setting several levels (e.g., 480p, 960p, and 1440p etc.) as a knob in works [8]–[10], smooth configuration adaption judiciously arbitrates the trade-off between accuracy and resource consumption.”
BackgroundMethodInfluentialEdge Computing Resource Management for Cross-Camera Video Analytics: Workload and Model Adaptation
Chen, H.-T., Chiang, Y., & Wei, H. (2024) · IEEE Access · doi:10.1109/access.2024.3355815 · Cited by 7
“Chameleon [23] adapted the configurations over time and utilized both spatial and temporal correlations to amortize the overhead of adaptation.”
BackgroundZhang, L., Zhu, H., Li, Y., Shen, J., & Guo, M. (2024) · Proceedings of the International Conference on Parallel Processing · doi:10.1145/3673038.3673081 · Cited by 7
Bayesian Online Learning for MEC Object Recognition Systems
Galanopoulos, A., Ayala‐Romero, J. A., Iosifidis, G., & Leith, D. J. (2020) · Globecom · doi:10.1109/globecom42002.2020.9322146 · Cited by 7
“Another option is pruning parts of the action space by limiting it to a number of competitive (high reward) actions [9], [25].”
BackgroundMethodThink before You Leap: Content-Aware Low-Cost Edge-Assisted Video Semantic Segmentation
Yan, M., Wang, Y., Xiao, X., Luo, Z., He, J., & Wang, W. (2023) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3581783.3613808 · Cited by 7
“Specifically, we compress the original videos into 18 variances, with QPs ranging from 20 to 30 and resolution scaling factors of 1, 0.75, and 0.5.”
BackgroundMethodInfluentialLogan: Loss-tolerant Live Video Analytics System
Yang, K., Jeong, M., Yi, J., Lee, J., Kyoungsoo, P., & Lee, Y. (2024) · ACM Annual International Conference on Mobile Computing and Networking (MobiCom) · doi:10.1145/3636534.3690695 · Cited by 7
Pontes, F. A., Schukat, M., & Curry, E. (2024) · IEEE Access · doi:10.1109/access.2024.3520807 · Cited by 7
“Similarly to other works in this field [8], [14], [53], an automatic annotation process was used for labelling the ground truth of the dataset.”
EdgeBooster: Edge-Assisted Real-Time Image Segmentation for the Mobile Web in WoT
Huang, Y., Qiao, X., Ren, P., Dustdar, S., & Chen, J. (2020) · IEEE Internet of Things Journal · doi:10.1109/jiot.2020.3038689 · Cited by 7
“Mobile Computing Offloading: Offloading computation from the mobile Web to the cloud or the edge server is a considerable way to enable continuous vision tasks [25], [26], [34], [48]–[50].”
BackgroundVision Paper: Towards Software-Defined Video Analytics with Cross-Camera Collaboration
Yi, J., Min, C., & Kawsar, F. (2021) · ACM Conference on Embedded Networked Sensor Systems (SenSys) · doi:10.1145/3485730.3493453 · Cited by 7
“Most of the existing works, however, have been focused on single-shot object detection tasks [5, 8, 19] and lacks considerations for such abstraction.”
BackgroundInfluentialVPPlus: Exploring the Potentials of Video Processing for Live Video Analytics at the Edge
Guo, J., Xia, S., & Peng, C. (2022) · International Workshop on Quality of Service · doi:10.1109/iwqos54832.2022.9812896 · Cited by 7
“• We have designed VPPlus to support a broader range of configuration parameters to tune on-device processing for [12], [22]–[24] No feedback √ 1 [10], [11], [26] Explicit fewer bytes and comparable analytical accuracy (§IV).”
BackgroundMethodInfluentialBasaras, P., Vasilopoulos, E., Magklaris, S., Katsaros, K. V., & Amditis, A. (2023) · IEEE Conference on Network Function Virtualization and Software Defined Networks (NFV-SDN) · doi:10.1109/nfv-sdn59219.2023.10329735 · Cited by 7
“Similarly, other studies focusing on reducing resource consumption (e.g., energy or compute) with little degradation in accuracy include those reported in [4], [12] or [15].”
BackgroundSupporting Video Queries on Zero-Streaming Cameras.Preprint
Xu, M., Xu, T., Liu, Y., Liu, X., Huang, G., & Lin, F. X. (2019) · arXiv (Cornell University) · Cited by 7
INFaaS: A Model-less Inference Serving SystemPreprint
Romero, F., Li, Q., Yadwadkar, N. J., & Kozyrakis, C. (2019) · arXiv (Cornell University) · Cited by 7
Righteous: Automatic Right-Sizing for Complex Edge Deployments
Rakshit, A., Reddy, S., Ramnath, R., Arora, A., & Boubin, J. G. (2024) · IEEE/ACM Symposium on Edge Computing (SEC) · doi:10.1109/sec62691.2024.00010 · Cited by 6
“Chameleon is a system for tuning video analytics applications to fit the form factor of available compute resources [29].”
Wang, S., Tan, X., Pan, Y., Li, O., Sun, M., Zheng, Q., & Yang, J. (2025) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2025.3577032 · Cited by 6
“Chameleon, proposed in [11], dynamically selects the best configurations for existing video analytics pipelines based on a heuristic hill-climbing solution.”
Huang, H., Liang, J., & Min, G. (2025) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2025.3586793 · Cited by 6
“To address this, Jiang et al. [16] developed Chameleon to reduce the configuration search space from exponential to linear through three key observations: temporal correlation, cross-camera correlation, and independence of configurations.”
Compass: SLO-Aware Query Planner for Compound AI Serving at Scale
Liu, B., Lin, W.-Y., Fang, M., Jiang, Y., & Lai, F. (2026) · Proceedings of the VLDB Endowment · doi:10.14778/3819518.3819524 · Cited by 6
Decentralized modular architecture for live video analytics at the edge
Rachuri, S. P., Bronzino, F., & Jain, S. (2021) · doi:10.1145/3477083.3480153 · Cited by 6
Rev: A Video Engine for Object Re-identification at the City Scale
Xu, T., Shen, K., Fu, Y. L., Shi, H., & Lin, F. X. (2022) · IEEE/ACM Symposium on Edge Computing (SEC) · doi:10.1109/sec54971.2022.00022 · Cited by 6
“There have been works optimizing video analytics with operator cascades [32], [16], [21], [56], and format tuning to trade accuracy for cost-efficiency [65], [25], [28], [67], [49].”
BackgroundRetina-U: A Two-Level Real-Time Analytics Framework for UHD Live Video Streaming
Zhang, W., Jing, Y., Zhang, Y., Lin, T., & Yan, J. (2024) · IEEE Transactions on Broadcasting · doi:10.1109/tbc.2023.3345646 · Cited by 6
“For example, Chameleon [39] adopts dynamic adjustments of video analytics configurations to trade off bandwidth utilization and inference accuracy.”
BackgroundWiseCam: A Systematic Approach to Intelligent Pan-Tilt Cameras for Moving Object Tracking
Jinlong, E., Han, F., He, L., Xu, W. X., Li, Z., Chai, Y., & Liu, Y. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3410645 · Cited by 6
“Specifically, Chameleon [35] dynamicallypicksthebestNNconfigurationtobalanceaccuracy and resource overhead.”
Incentive Mechanism for Resource Trading in Video Analytic Services Using Reinforcement Learning
He, N., Yang, S., Li, F., Zhu, L., Sun, L., Chen, X., & Fu, X. (2024) · IEEE Transactions on Services Computing · doi:10.1109/tsc.2024.3424220 · Cited by 6
“Jiang et al. [28] showcase a video analytics system that dynamically selects optimal configurations for existing DNNs-based video analytics pipelines.”
VideoJam: Self-Balancing Architecture for Live Video Analytics
Faye, Y., Faticanti, F., Jain, S., & Bronzino, F. (2024) · IEEE/ACM Symposium on Edge Computing (SEC) · doi:10.1109/sec62691.2024.00020 · Cited by 6
JSFBA: Joint Segment and Frame Bitrate Adaptation for Real-Time Video Analytics
Wang, Y., Cheng, S., Gao, N., Bi, T., & Jiang, T. (2025) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2025.3526867 · Cited by 6
“There have been extensive studies on the effect of video configuration on video analytics performance [19], [20], [21], and the results indicate that adjusting the video configura-tion can reduce bandwidth requirements while maintaining analytics accuracy.”
Xarchakos, I., & Koudas, N. (2021) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2021.3094997 · Cited by 6
“[20], [23], [36] present system approaches to query processing across video streams and are concerned with scalability, resource utilization and storage aspects.”
Background
Metadata
- Authors
- Jiang, J., Ananthanarayanan, G., Bodik, P., Sen, S., & Stoica, I.
- Affiliations
- University of Chicago (US)
- Microsoft (United States) (US)
- Microsoft Research
- University of California, Berkeley (US)
- Venue
- Proceedings of the 2018 Conference of the ACM Special Interest Group on Data Communication
- Publisher
- ACM
- Published
- 2018-08-07
- Pages
- 253-266
- OpenAlex
- W2887117815
- Cited by
- 467
- References
- 25
- Citations by year
- 2026: 33 · 2025: 58 · 2024: 69 · 2023: 74 · 2022: 70 · 2021: 69 · 2020: 50 · 2019: 38 · 2018: 6
- FWCI
- 12.28
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- ACM SIGCOMM Conference (SIGCOMM) (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Video Surveillance and Tracking Methods
- Physical Sciences › Computer Science › Information Systems › Advanced Database Systems and Queries
- Keywords
- video analysis, deep convolutional neural networks, object detection, chameleons
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.