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 101–200 of 575
Xu, R., Lee, J., Wang, P., Bagchi, S., Li, Y., & Chaterji, S. (2022) · European Conference on Computer Systems · doi:10.1145/3492321.3519577 · Cited by 35
Serving and Optimizing Machine Learning Workflows on Heterogeneous Infrastructures
Wu, Y., Lentz, M., Zhuo, D., & Lu, Y. (2022) · Proceedings of the VLDB Endowment · doi:10.14778/3570690.3570692 · Cited by 35
Edge-Assisted Real-Time Video Analytics With Spatial–Temporal Redundancy Suppression
Wang, Z., He, X., Zhang, Z., Zhang, Y., Cao, Z. Z., Cheng, W., Wang, W., & Cui, Y. (2022) · IEEE Internet of Things Journal · doi:10.1109/jiot.2022.3224750 · Cited by 34
“3) Some other researchers adapted configurations (e.g., frame rate, frame resolution, and deep learning model) to optimize a given goal [11], [12], [13], [23], [24], [25], [26].”
BackgroundLeveraging Cloud Computing to Make Autonomous Vehicles Safer
Schafhalter, P., Kalra, S., Xu, L., Gonzalez, J. E., & Stoica, I. G. (2023) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros55552.2023.10341821 · Cited by 34
“In addition to deciding whether to offload the computation of the messages to the cloud based on its contents [46,70], the handler functions can use the logical timestamps to enforce simple policies such as sending non-consecutive inputs to the cloud to reduce bandwidth usage, or alternating…”
BackgroundSelf-citationEdge-Based Video Surveillance With Graph-Assisted Reinforcement Learning in Smart Construction
Ming, Z., Jin-shen, C., Cui, L., Yang, S., Pan, Y. F., Xiao, W., & Zhou, L. (2021) · IEEE Internet of Things Journal · doi:10.1109/jiot.2021.3090513 · Cited by 33
Towards Transmission-Friendly and Robust CNN Models over Cloud and Device
Ding, C., Lu, Z., Juefei-Xu, F., Boddeti, V. N., Li, Y., & Cao, J. (2022) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2022.3186496 · Cited by 33
“[12] proposed a video analysis controller based on cloud and deep neural networks.”
BackgroundMethodVisual Road: A Video Data Management Benchmark
Haynes, B. M., Mazumdar, A., Bałazińska, M., Ceze, L., & Cheung, A. (2019) · ACM SIGMOD International Conference on Management of Data (SIGMOD) · doi:10.1145/3299869.3324955 · Cited by 33
“Visual Road is designed to be implementable across a wide variety of VDBMS architectures, including those that perform video querying at scale (e.g., Scanner [44], Optasia [35], Chameleon [31]), operate on emerging forms of video data (e.g., LightDB [20]), and perform deep learning inference (e.g., NoScope [33], Blaze…”
BackgroundInfluentialVideoPipe: Building Video Stream Processing Pipelines at the Edge
Salehe, M., Hu, Z., Mortazavi, S. H., Mohomed, I., & Capes, T. (2019) · International Middleware Conference · doi:10.1145/3366626.3368131 · Cited by 33
A Case for Managed and Model-less Inference Serving
Yadwadkar, N. J., Romero, F., Li, Q., & Kozyrakis, C. (2019) · Proceedings of the Workshop on Hot Topics in Operating Systems · doi:10.1145/3317550.3321443 · Cited by 31
“Recent trends in analytics [11, 24, 29] indicate a considerable increase in the use of predictions from machine learning (ML) models, especially neural networks, by various applications.”
BackgroundBoosting Edge Intelligence With Collaborative Cross-Edge Analytics
Jin, H., Jia, L., & Zhou, Z. (2020) · IEEE Internet of Things Journal · doi:10.1109/jiot.2020.3034891 · Cited by 31
“On one hand, to reduce the latency and/or resource consumption of the model inference phase of edge intelligence, various approaches, such as model compression [9], model selection [10], and collaborative model execution [11] have been proposed.”
MethodDeep Learning Video Analytics Through Online Learning Based Edge Computing
Liu, H., & Cao, G. (2022) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2022.3164598 · Cited by 31
“Researchers have proposed various offloading frameworks 114 for video analytics in the past several years [7], [8], [10]– 115 [12].”
BackgroundFrameHopper: Selective Processing of Video Frames in Detection-driven Real-Time Video Analytics
Arefeen, M. A., Nimi, S. T., & Uddin, M. Y. S. (2022) · International Conference on Distributed Computing in Smart Systems and the Internet of Things (DCOSS-IoT) · doi:10.1109/dcoss54816.2022.00033 · Cited by 31
A Serverless Cloud-Fog Platform for DNN-Based Video Analytics with Incremental LearningPreprint
Zhang, H., Shen, M., Huang, Y., Wen, Y., Luo, Y., Gao, G., & Guan, K. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.03012 · Cited by 31
“Second, most of these works over-rely on the performances of the state-of-the-art models (e.g., FasterRCNN101) to guide their design options [18].”
BackgroundMethodInfluentialEdgeBOL: automating energy-savings for mobile edge AI
Ayala‐Romero, J. A., García‐Saavedra, A., Costa‐Pérez, X., & Iosifidis, G. (2021) · ACM International Conference on emerging Networking EXperiments and Technologies (CoNEXT) · doi:10.1145/3485983.3494849 · Cited by 30
A3D: Adaptive, Accurate, and Autonomous Navigation for Edge-Assisted Drones
Zeng, L., Chen, H., Feng, D., Zhang, X., & Chen, X. (2023) · IEEE/ACM Transactions on Networking · doi:10.1109/tnet.2023.3297876 · Cited by 30
“Existing systems (e.g., [16], [18]) usually achieve dynamic input resolution by loading a group of models that accept different input sizes and switching the execution target at runtime, which may take a large volume of memory and bring model switching overhead.”
BackgroundMethodKim, W., & Youn, C. (2020) · IEEE Access · doi:10.1109/access.2020.3004571 · Cited by 30
“One of the promising solutions to address this issue is adapting configurations optimally in the video analytics system [10], [11].”
BackgroundMethodInfluentialExSample: Efficient Searches on Video Repositories through Adaptive Sampling
Moll, O., Bastani, F., Madden, S. R., Stonebraker, M., Gadepally, V. N., & Kraska, T. (2022) · IEEE International Conference on Data Engineering (ICDE) · doi:10.1109/icde53745.2022.00266 · Cited by 31
“These datasets are used in the evaluation of related work on optimizing (other) types of video queries on static camera datasets [6], [10], [21].”
MethodAdaptive Computation Offloading for Mobile Augmented Reality
Ren, J., Gao, L. C., Wang, X., Ma, M., Qiu, G., Wang, H., Zheng, J., & Wang, Z. (2021) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3494958 · Cited by 29
“To minimize resource utilization while meeting the user-specified accuracy requirement, Chameleon [38] and VideoStorm [115] reduce the backend computation costs by profiling different configurations of pipeline knobs (e.g., video resolution, frame sampling rate, etc.).”
MethodPacketGame: Multi-Stream Packet Gating for Concurrent Video Inference at Scale
Yuan, M., Zhang, L., You, X., & Li, X. (2023) · ACM SIGCOMM Conference (SIGCOMM) · doi:10.1145/3603269.3604825 · Cited by 29
“A typical video inference pipeline [19, 20, 31, 34, 53, 55] (see Fig.”
MethodResource Efficient Edge Computing Infrastructure for Video Surveillance
Kumar, P. P., Pal, A., & Kant, K. (2021) · IEEE Transactions on Sustainable Computing · doi:10.1109/tsusc.2021.3064245 · Cited by 28
“In [12], authors have found that the frame sampling rate independently impacts accuracy.”
BackgroundOn unifying deep learning and edge computing for human motion analysis in exergames development
Pardos, A., Menychtas, A., & Maglogiannis, I. G. (2021) · Neural Computing and Applications · doi:10.1007/s00521-021-06181-6 · Cited by 28
“This is proven by the increasing number of solution that are migrated from cloud to the edge due to the great potential of edge-based [21, 52] systems.”
BackgroundWang, X., Gao, G., Wu, X., Lyu, Y., & Wu, W. (2022) · International Workshop on Network and Operating Systems Support for Digital Audio and Video (NOSSDAV) · doi:10.1145/3534088.3534352 · Cited by 28
“Chameleon [4], DeepDecision [9], and A2 [5] studied the conguration of video analytics pipelines to optimize delay and accuracy.”
BackgroundJiang, S., Shuai, X., & Xing, G. (2024) · ACM/IEEE International Conference on Information Processing in Sensor Networks (IPSN) · doi:10.1109/ipsn61024.2024.00007 · Cited by 28
“Consequently, prior works [21, 60] have employed dynamic resolution tuning during inference.”
InfluentialYadav, P., Salwala, D., & Curry, E. (2021) · IEEE Internet of Things Journal · doi:10.1109/jiot.2021.3075336 · Cited by 28
“VideoStorm [45] and Chameleon [44] processes video streams by tuning knobs as per resource quality tradeoffs on large clusters with no focus on edge deployments, windows and state management.”
BackgroundMethodLlama: A Heterogeneous & Serverless Framework for Auto-Tuning Video Analytics PipelinesPreprint
Romero, F., Zhao, M., Yadwadkar, N. J., & Kozyrakis, C. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.01887 · Cited by 28
Batch Adaptative Streaming for Video Analytics
Zhang, L., Zhang, Y., Wu, X., Wang, F., Cui, L., Wang, Z., & Liu, J. (2022) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom48880.2022.9796853 · Cited by 27
“Chameleon [19] applies new configuration for the video analytics pipeline based on periodical profiling.”
BackgroundMethodInfluentialExtract-Transform-Load for Video Streams
Kossmann, F., Wu, Z., Lai, E. Y., Tatbul, N., Cao, L., Kraska, T., & Madden, S. R. (2023) · Proceedings of the VLDB Endowment · doi:10.14778/3598581.3598600 · Cited by 27
EQUI-VOCAL: Synthesizing Queries for Compositional Video Events from Limited User Interactions
Zhang, E., Daum, M., He, D., Haynes, B. M., Krishna, R., & Bałazińska, M. (2023) · Proceedings of the VLDB Endowment · doi:10.14778/3611479.3611482 · Cited by 27
“Many recent VDBMSs have been proposed and focus on a wide range of data management challenges, including fast inference over videos [3, 5, 34, 44, 47], storage optimization [18, 25, 64], efficient dataflow processing [49], preprocessing and indexing [6, 28, 36], exploration and organization [19], privacy [9], and tuni…”
BackgroundJoint Model and Data Adaptation for Cloud Inference Serving
Jiang, J., Luo, Z., Hu, C., He, Z., Wang, Z., Xia, S., & Wu, C. (2021) · IEEE Real-Time Systems Symposium (RTSS) · doi:10.1109/rtss52674.2021.00034 · Cited by 27
Self-citationWu, D., Zhang, D., Zhang, M., Zhang, R., Wang, F., & Cui, S. R. (2023) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2023.3327097 · Cited by 26
“Previous works usually use profiling-based strategy to seek the best configurations, i.e., searching the possible combination of knobs on a small portion of a given video clip and applying the derived best one to the rest portion [7], [8].”
BackgroundMethodInfluentialDeepHome: Distributed Inference with Heterogeneous Devices in the Edge
Hu, Z., Tarakji, A. B., Raheja, V., Phillips, C., Wang, T., & Mohomed, I. (2019) · doi:10.1145/3325413.3329787 · Cited by 26
“Apart from applications on image inference, data analytics on video data is another useful application in the edge using machine learning [6, 11, 20].”
MethodLive Video Analytics with FPGA-based Smart Cameras
Wang, S., Zhang, C., Shu, Y., & Liu, Y. (2019) · doi:10.1145/3349614.3356027 · Cited by 26
Gupta, A. K., & Prabhat, P. (2022) · Complex & Intelligent Systems · doi:10.1007/s40747-022-00783-w · Cited by 26
“Chameleon [22] is a video-surveillance system that dynamically adjusts the video stream configurations based on video segment profiles intending to optimize resource utilization and increase accuracy.”
BackgroundMulti-View Scheduling of Onboard Live Video Analytics to Minimize Frame Processing Latency
Liu, S., Wang, T., Guo, H., Fu, X., David, P., Wigness, M., Misra, A., & Abdelzaher, T. (2022) · IEEE International Conference on Distributed Computing Systems (ICDCS) · doi:10.1109/icdcs54860.2022.00055 · Cited by 26
“The key idea in this thread is to minimize the amount [3], [26], [28], resolutions [27], or the regions [16] of frames to be transmitted, because the network bandwidth is the main bottleneck in this pipeline.”
BackgroundEdgeSharing: Edge Assisted Real-time Localization and Object Sharing in Urban Streets
Liu, L., & Gruteser, M. (2021) · IEEE INFOCOM 2021 - IEEE Conference on Computer Communications · doi:10.1109/infocom42981.2021.9488830 · Cited by 25
“VideoStorm [32] and Chameleon [33] achieve higher accuracy video analytics with the same amount of computational resources on the cloud by adapting the video configurations.”
BackgroundVaVLM: Toward Efficient Edge-Cloud Video Analytics With Vision-Language Models
Zhang, Y., Wang, H., Bai, Q., Liang, H., Zhu, P., Muntean, G., & Li, Q. (2025) · IEEE Transactions on Broadcasting · doi:10.1109/tbc.2025.3549983 · Cited by 25
“…models suitable for deployment on edge devices using techniques like NAS [14], [15], [47], [48], [49]; ( iii ) Resource optimization-based approaches, which focus on improving overall system performance by optimizing the use of bandwidth and computational resources [18], [50], [51], [52].”
Anomalous Vehicle Recognition in Smart Urban Traffic Monitoring as an Edge Service
Chen, N., & Chen, Y. (2022) · Future Internet · doi:10.3390/fi14020054 · Cited by 25
“In order to enable deep learning at the edge, researchers explore various approaches [37]: (1) reducing the latency of executing deep learning on resource constrained devices by customized model design [38–40], model compression [41–43], and hardware-based accelerators [44–46]; (2) off-loading computing intensive task…”
BackgroundDai, X., Zhang, Z., Yang, P., Xu, Y., Liu, X., & Lui, J. C. S. (2024) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3664647.3681269 · Cited by 25
Perceptual Compression for Video Storage and Processing Systems
Mazumdar, A., Haynes, B. M., Bałazińska, M., Ceze, L., Cheung, A., & Oskin, M. (2019) · ACM Symposium on Cloud Computing (SoCC) · doi:10.1145/3357223.3362725 · Cited by 25
Multi-user Edge-assisted Video Analytics Task Offloading Game based on Deep Reinforcement Learning
Chen, Y., Zhang, S., Xiao, M., Qian, Z., Wu, J., & Lu, S. (2020) · IEEE International Conference on Parallel and Distributed Systems (ICPADS) · doi:10.1109/icpads51040.2020.00044 · Cited by 24
“Same as the existing works [7], [8], we refer to the combination of frame rate and resolution as a configuration, and different configurations lead to different accuracies and resource consumptions.”
BackgroundMethodTowards Timely Video Analytics Services at the Network Edge
Li, X., Zhang, S., Huang, Y., Ma, X., Wang, Z., & Luo, H. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3376769 · Cited by 24
“For example, in [7], the temporal and spatial correlations in video contents are carefully utilized to help choose video configurations to achieve a higher accuracy.”
Adapting Neural Networks at Runtime: Current Trends in At-Runtime Optimizations for Deep Learning
Sponner, M., Waschneck, B., & Kumar, A. S. (2024) · ACM Computing Surveys · doi:10.1145/3657283 · Cited by 24
Deep Learning Video Analytics on Edge Computing Devices
Tan, T., & Cao, G. (2021) · IEEE International Conference on Sensing, Communication, and Networking (SECON) · doi:10.1109/secon52354.2021.9491614 · Cited by 24
“Chameleon [14] achieve a better tradeoff between accuracy and processing time by dynamically changing the configuration of DNN used in video analytics.”
MethodA Feedback-Driven DNN Inference Acceleration System for Edge-Assisted Video Analytics
Lv, X., Wang, Q., Chen, Y., & Jin, H. (2023) · IEEE Transactions on Computers · doi:10.1109/tc.2023.3275094 · Cited by 23
“And some systems try to first use a small lightweight model locally or process the frame with low resolution on the edge to obtain large objects of the frame, and then judge the area containing small objects based on the detection confidence, last use the large accurate model for the detection of small objects [21], […”
BackgroundRT3C: Real-Time Crowd Counting in Multi-Scene Video Streams via Cloud-Edge-Device Collaboration
Wang, R., Hao, Y., Miao, Y., Hu, L., & Chen, M. (2024) · IEEE Transactions on Services Computing · doi:10.1109/tsc.2024.3377156 · Cited by 23
“Many methods have been proposed to meet the demands of latency and performance in the real-time video analysis systems [11], [18], [19], [20], [21], [22].”
MethodComprehensive and Efficient Data Labeling via Adaptive Model Scheduling
Yuan, M., Zhang, L., Li, X., & Xiong, H. (2020) · IEEE International Conference on Data Engineering (ICDE) · doi:10.1109/icde48307.2020.00188 · Cited by 23
“The other stream of work focus on accelerating the model execution by designing a variety of methods, including model compression via parameter pruning & sharing [4], network architecture optimization [29] and adaptive model configuration [12].”
MethodNeuLens: spatial-based dynamic acceleration of convolutional neural networks on edge
Hou, X., Guan, Y., & Han, T. (2022) · ACM Annual International Conference on Mobile Computing and Networking (MobiCom) · doi:10.1145/3495243.3560528 · Cited by 23
“Recent works on continuous object detection [5, 36, 47, 84] exploit the temporal correlation between consecutive frames and reduce the overall computation of video streams by combining model detection with tracking algorithms [47, 84].”
BackgroundDeepLens: Towards a Visual Data Management SystemPreprint
Krishnan, S., Dziedzic, A., & Elmore, A. J. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1812.07607 · Cited by 23
“Answering such queries at scale is a formidable computer systems challenge with significant recent interest from both industry and academia [3,4,6, 10,11,13,15,21,22,27].”
BackgroundInfluentialEnhancing Video Analytics Accuracy via Real-time Automated Camera Parameter Tuning
Paul, S., Rao, K., Coviello, G., Sankaradas, M., Po, O., Hu, Y., & Chakradhar, S. (2022) · ACM Conference on Embedded Networked Sensor Systems (SenSys) · doi:10.1145/3560905.3568527 · Cited by 22
ExSample: Efficient Searches on Video Repositories through Adaptive Sampling
Moll, O., Bastani, F., Madden, S. R., Stonebraker, M., Gadepally, V. N., & Kraska, T. (2022) · IEEE International Conference on Data Engineering (ICDE) · doi:10.1109/icde53745.2022.00275 · Cited by 22
Evaluating Temporal Queries Over Video Feeds
Chen, Y., Yu, X., Koudas, N., & Yu, Z. (2021) · ACM SIGMOD International Conference on Management of Data (SIGMOD) · doi:10.1145/3448016.3452803 · Cited by 22
“Other related works [18, 19, 28, 42] present system approaches to query processing across video streams and are concerned with scalability, resource utilization and storage aspects.”
BackgroundArtificial intelligence in edge devices
Singh, A., Saini, K., Nagar, V., Aseri, V., Sankhla, M. S., Pandit, P. P., & Chopade, R. L. (2022) · Advances in computers · doi:10.1016/bs.adcom.2022.02.013 · Cited by 22
CrossVision: Real-Time On-Camera Video Analysis via Common RoI Load Balancing
Zhang, L., Lu, Z., Song, L., & Xu, J. (2023) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2023.3301391 · Cited by 21
“2, two pipelines are commonly used in existing live video analytics systems [15].”
BackgroundZhuang, W., Xing, F., & Lu, Y. (2024) · Sensors · doi:10.3390/s24072070 · Cited by 22
“This approach allows for the proximity of computation and data storage functions to inspection sites, resulting in lower latency and higher real-time capabilities [13,14].”
BackgroundAccelerated Neural Enhancement for Video Analytics With Video Quality Adaptation
Mi, L., Yuan, T., Wang, W., Dai, H., Sun, L., Zheng, J., Chen, G., & Fu, X. (2024) · IEEE/ACM Transactions on Networking · doi:10.1109/tnet.2024.3375108 · Cited by 21
“On the one hand, modern cameras can generate video chunks in different resolutions because they can achieve self-adaptive encoding by dynamically adjusting resolutions to accommodate network fluctuations and object size, as demonstrated in [25] and [26].”
BackgroundMethodARISE: High-Capacity AR Offloading Inference Serving via Proactive Scheduling
Kong, Z. J., Xu, Q., & Hu, Y. (2024) · International Conference on Mobile Systems, Applications, and Services · doi:10.1145/3643832.3661894 · Cited by 21
“…using a large set of emulated AR clients and a small-scale testbed of 10 phones show that compared to Clipper-like [21] and Chameleon-like [34] systems, ARISE provides significantly higher capacity in serving AR clients, by up to 5.2x and 6.9x for the depth estimation task and by 1.9x and…”
BackgroundMethodInfluentialVisage: enabling timely analytics for drone imagery
Jha, S., Li, Y., Noghabi, S. A., Ranganathan, V. N., Kumar, P., Nelson, A., Toelle, M., Sinha, S. N., Chandra, R., & Badam, A. (2021) · ACM Annual International Conference on Mobile Computing and Networking (MobiCom) · doi:10.1145/3447993.3483273 · Cited by 21
On the scalability of Big Data Cyber Security Analytics systems
Ullah, F., & Babar, M. A. (2021) · Journal of Network and Computer Applications · doi:10.1016/j.jnca.2021.103294 · Cited by 21
“[90] proposed a parameter-driven adaptation approach that uses the temporal and spatial correlations among characteristics (such as size and velocity of objects) for finding the best set of configuration parameters for a convolutional neural network employed in a video analytics system.”
BackgroundMethodTurbo: Opportunistic Enhancement for Edge Video Analytics
Lu, Y., Jiang, S., Cao, T., & Shu, Y. (2022) · ACM Conference on Embedded Networked Sensor Systems (SenSys) · doi:10.1145/3560905.3568501 · Cited by 21
“The first [4, 5, 20, 21, 25, 26, 30, 33, 44] is based on temporal consistency of videos and remove similar frames by inter-frame difference [5, 26].”
BackgroundMethodInfluentialBiSwift: Bandwidth Orchestrator for Multi-Stream Video Analytics on Edge
Sun, L., Wang, W., Yuan, T., Mi, L., Dai, H., Liu, Y., & Fu, X. (2024) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom52122.2024.10621392 · Cited by 20
Argus: Enabling Cross-Camera Collaboration for Video Analytics on Distributed Smart Cameras
Yi, J., Acer, U. G., Kawsar, F., & Min, C. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3459409 · Cited by 20
“The other direction for efficient processing is DNN inference scheduling from multiple video streams on the GPU cluster [17], [18], [70], DNN merging for memory optimisation [71], privacy-aware video analytics [72], [73], and resource-efficient continual learning [74], [75].”
SmartFilter: An Edge System for Real-Time Application-Guided Video Frames Filtering
Tchaye-Kondi, J., Zhai, Y., Shen, J., Dong, L., & Zhu, L. (2022) · IEEE Internet of Things Journal · doi:10.1109/jiot.2022.3188518 · Cited by 20
Stylianopoulos, K., Merluzzi, M., Lorenzo, P. D., & Alexandropoulos, G. C. (2023) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49357.2023.10095112 · Cited by 20
Networked Cameras Are the New Big Data Clusters
Jiang, J., Zhou, Y., Ananthanarayanan, G., Shu, Y., & Chien, A. A. (2019) · doi:10.1145/3349614.3356026 · Cited by 21
Self-citationPrivid: Practical, Privacy-Preserving Video Analytics QueriesPreprint
Cangialosi, F., Agarwal, N., Arun, V., Jiang, J., Narayana, S., Sarwate, A. D., & Netravali, R. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.12083 · Cited by 20
“tics (both live and retrospective) at a massive scale across entire networks of cameras. While these trends enable a variety of importantapplications[2,10,12,13]andfuelmuchworkinthe systems community [24,27,34,37,38,41,42,47,63], they also enable privacy intrusions at an unprecedented level [7,53]. Asaconcreteexample,…”
BackgroundSelf-citationNNFacet: Splitting Neural Network for Concurrent Smart Sensors
Chen, J., Le, D. V., Tan, R., & Ho, D. (2023) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2023.3238342 · Cited by 19
“The video profiling method proposed in [29] can be used to evaluate the temporal correlation between the consecutive frames.”
MethodREACT: Streaming Video Analytics On The Edge With Asynchronous Cloud Support
Ghosh, A., Iyengar, S., Lee, S., Rathore, A., & Padmanabhan, V. N. (2023) · IEEE/ACM International Conference on Internet-of-Things Design and Implementation (IoTDI) · doi:10.1145/3576842.3582385 · Cited by 19
MOSAIC: Spatially-Multiplexed Edge AI Optimization over Multiple Concurrent Video Sensing Streams
Gokarn, I., Sabbella, H., Hu, Y., Abdelzaher, T., & Misra, A. (2023) · ACM Multimedia Systems Conference (MMSys) · doi:10.1145/3587819.3590986 · Cited by 19
VisualWorldDB: A DBMS for the Visual World.
Haynes, B. M., Daum, M., Mazumdar, A., Bałazińska, M., Cheung, A., & Ceze, L. (2020) · Conference on Innovative Data Systems Research · Cited by 19
Palleon: A Runtime System for Efficient Video Processing toward Dynamic Class Skew
Feng, B., Wang, Y., Li, G., Xie, Y., & Ding, Y. (2021) · USENIX Annual Technical Conference · Cited by 19
AMVP: Adaptive CNN-based Multitask Video Processing on Mobile Stream Processing Platforms
Chao, M., Stoleru, R., Jin, L., Yao, S., Maurice, M., & Blalock, R. (2020) · IEEE/ACM Symposium on Edge Computing (SEC) · doi:10.1109/sec50012.2020.00015 · Cited by 19
“For the future work, we consider applying similar accuracy metrics as in Chameleon [49], which includes precision, recall and F1 score.”
TOD: Transprecise Object Detection to Maximise Real-Time Accuracy on the Edge
Lee, J., Varghese, B., Woods, R., & Vandierendonck, H. (2021) · IEEE International Conference on Fog and Edge Computing (ICFEC) · doi:10.1109/icfec51620.2021.00015 · Cited by 19
“Comparison to Related Work: Rather than seeking the optimal FPS a DNN can handle in [3], TOD automatically chooses a slower network if a higher FPS is not required.”
BackgroundMethodInfluentialINFaaS: Managed & Model-less Inference ServingPreprint
Romero, F., Li, Q., Yadwadkar, N. J., & Kozyrakis, C. (2019) · arXiv (Cornell University) · Cited by 19
Towards Performance Clarity of Edge Video AnalyticsPreprint
Xiao, Z., Xia, Z., Zheng, H.-T., Zhao, B. Y., & Jiang, J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2105.08694 · Cited by 19
“Videostorm [94] and Chameleon [54] uniformly sample frames, downsize ( ∼ 5 Mbps) costs $2K/day for AT&T 4G LTE network ($50 for 30GB data before the speed drops to a measly 128kbps [2]).”
BackgroundInfluentialSelf-citationTop-K Deep Video Analytics: A Probabilistic Approach
Lai, Z., Han, C., Liu, C., Zhang, P., Lo, E., & Kao, B. (2021) · ACM SIGMOD International Conference on Management of Data (SIGMOD) · doi:10.1145/3448016.3452786 · Cited by 18
PATCH: A Plug-in Framework of Non-blocking Inference for Distributed Multimodal System
Wang, J., Wang, G., Zhang, X., Liu, L., Zeng, H., Xiao, L., Cao, Z., Gu, L., & Li, T. (2023) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3610885 · Cited by 18
“Low-latency Deep Learning in IoT Systems Prior work mainly focuses on optimizing resource-accuracy trade-offs to reduce latency in IoT systems [9, 28, 29, 41, 86].”
BackgroundAdaptive Network Configuration for Efficient and Accurate Neural Video Inference
Yang, P., Cheng, Y., Zhang, N., Cheng, Q., & Yu, L. (2023) · IEEE Transactions on Cognitive Communications and Networking · doi:10.1109/tccn.2023.3320879 · Cited by 18
“Generally, the optimal configuration that requires the least computation resources with the accuracy guarantee varies over time [1], [30].”
BackgroundMethodInfluentialDai, P., Chao, Y., Wu, X. Y., Liu, K., & Guo, S. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3418608 · Cited by 18
“This approach aligns with the best practices established in prior studies [17], [29].”
ResultTHIA: Accelerating Video Analytics using Early Inference and Fine-Grained Query PlanningPreprint
Cao, J., Hadidi, R., Arulraj, J. J. P., & Kim, H. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.08481 · Cited by 18
“The authors of Chameleon [9] observe that an appropriate query plan is critical to gain high performance and accuracy.”
BackgroundWang, H., Li, T., Zhang, M., Li, Q., Cui, H., Jiang, Y., & Yuan, Z. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3465434 · Cited by 17
“Traditionally, this involves tuning configuration parameters across three dimensions: temporal (frame sampling rate) [11]–[13], spatial (video quality compression) [14], and model [15].”
InfluentialSoudain: Online Adaptive Profile Configuration for Real-time Video Analytics
Wu, K., Jin, Y., Miao, W., Zeng, Z., Qian, Z., Wang, J., Zhou, M., & Cao, T. (2021) · International Workshop on Quality of Service · doi:10.1109/iwqos52092.2021.9521328 · Cited by 17
TURBO: Utility-Aware Bandwidth Allocation for Cloud-Augmented Autonomous Control
Schafhalter, P., Krentsel, A., Wei, H., Gonzalez, J. E., Ratnasamy, S., Shenker, S., & Stoica, I. (2025) · DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · doi:10.4230/oasics.nines.2026.18 · Cited by 17
“Several systems reconfigure the ML inference pipeline to react to changes in the input content [147, 141, 142, 57, 79] and to maximize real-time accuracy metrics [72, 108].”
Self-citationLive Migration of Video Analytics Applications in Edge Computing
Rong, C., Wang, J. H., Wang, J., Zhou, Y., & Zhang, J. (2023) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2023.3246539 · Cited by 16
“An edge platform may change the placement of applications among edge servers (or between the edge server and the cloud) to improve resource efficiency [4], [5], [6], [7], [8], [9].”
BackgroundMethodInference serving with end-to-end latency SLOs over dynamic edge networks
Nigade, V., Bauszat, P., Bal, H. E., & Wang, L. (2024) · Real-Time Systems · doi:10.1007/s11241-024-09418-4 · Cited by 16
Better Never Than Late: Timely Edge Video Analytics Over the Air
Nigade, V., Winder, R., Bal, H. E., & Wang, L. (2021) · ACM Conference on Embedded Networked Sensor Systems (SenSys) · doi:10.1145/3485730.3493446 · Cited by 16
DeepScale: Online Frame Size Adaptation for Multi-object Tracking on Smart Cameras and Edge Servers
Nalaie, K., Xu, R., & Zheng, R. (2022) · IEEE/ACM International Conference on Internet-of-Things Design and Implementation (IoTDI) · doi:10.1109/iotdi54339.2022.00010 · Cited by 16
MLink: Linking Black-Box Models for Collaborative Multi-Model Inference
Yuan, M., Zhang, L., & Li, X. (2022) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v36i9.21180 · Cited by 16
“Adaptive configuration (Jiang et al. 2018) and multi-model scheduling (Yuan et al. 2020a) were proposed to make inference workloads adaptive to the dynamics of input content.”
Characterizing Distributed Mobile Augmented Reality Applications at the Edge
Bartolomeo, G., Cao, J., Su, X. J., & Mohan, N. (2023) · doi:10.1145/3624354.3630584 · Cited by 16
Knowledge Distillation for Mobile Edge Computation OffloadingPreprint
Chen, H., Zeng, L., Yu, S., & Chen, X. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2004.04366 · Cited by 16
Live video analytics as a service
Apostolo, G. H., Bauszat, P., Nigade, V., Bal, H. E., & Wang, L. (2022) · Workshop on Machine Learning and Systems (EuroMLSys) · doi:10.1145/3517207.3526973 · Cited by 15
RALF: Accuracy-Aware Scheduling for Feature Store Maintenance
Wooders, S., Mo, X., Narang, A., Lin, K., Stoica, I. G., Hellerstein, J. M., Crooks, N., & Gonzalez, J. E. (2023) · Proceedings of the VLDB Endowment · doi:10.14778/3632093.3632116 · Cited by 15
“Systems that do consider featuriza-tion either focus on making use of cheaper featurization functions which can be approximated without affecting prediction [32], or target specific application use cases such as video analytics [15, 27, 38].”
BackgroundSelf-citation[Solution] IPA: Inference Pipeline Adaptation to achieve high accuracy and cost-efficiency
Ghafouri, S., Razavi, K., Salmani, M., Sanaee, A., Lorido-Botrán, T., Wang, L., Doyle, J., & Jamshidi, P. (2024) · Journal of Systems Research · doi:10.5070/sr34163500 · Cited by 15
“Multi-stage inference serving : Several approaches have been proposed in previous research to improve inference performance metrics in multistage inference serving systems [7, 25, 37, 41, 42, 44, 47, 51, 58, 59, 61, 65, 69] since changing one model’s configuration affects subsequent steps.”
Learning for Crowdsourcing: Online Dispatch for Video Analytics with Guarantee
Chen, Y., Zhang, S., Jin, Y., Qian, Z., Xiao, M., Chen, N., & Ma, Z. (2022) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom48880.2022.9796960 · Cited by 15
AdaDrone: Quality of Navigation Based Neural Adaptive Scheduling for Edge-Assisted Drones
Chen, H., Zeng, L., Zhang, X., & Chen, X. (2022) · IEEE International Conference on Distributed Computing Systems (ICDCS) · doi:10.1109/icdcs54860.2022.00059 · Cited by 15
“Existing systems [14], [16] usually achieve dynamic input resolution by loading a group of models that accept different input sizes and switching the execution target at runtime, which may bring model switching overhead.”
BackgroundMethodSiEVE: Semantically Encoded Video Analytics on Edge and Cloud
Elgamal, T., Shi, S., Gupta, V., Jana, R., & Nahrstedt, K. (2020) · IEEE International Conference on Distributed Computing Systems (ICDCS) · doi:10.1109/icdcs47774.2020.00182 · Cited by 15
“Most of the existing object detection [4] [5] systems for live video streams do not leverage the 3-tier architecture, and instead they focus on a 2-tier architecture where the encoded video is sent from the camera/smartphone to either a remote cloud server or a private edge server.”
AdaPyramid: Adaptive Pyramid for Accelerating High-Resolution Object Detection on Edge Devices
Shi, X., Zhang, S., Wu, J., Chen, N., Cheng, K., Liang, Y., & Lu, S. (2023) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2023.3343448 · Cited by 14
“Therefore, common practices [30], [31] often rely on offline profiles for the configurations.”
BackgroundCACTUS: Dynamically Switchable Context-aware micro-Classifiers for Efficient IoT Inference
Rastikerdar, M. M., Huang, J., Fang, S., Guan, H., & Ganesan, D. (2024) · International Conference on Mobile Systems, Applications, and Services · doi:10.1145/3643832.3661888 · Cited by 14
“The first set of approaches [26, 33, 35] are query-based and assume scenarios are known offline (static context).”
BackgroundVisFlow: Adaptive Content-Aware Video Analytics on Collaborative Cameras
Yan, Y., Zhang, S., Wang, X., Chen, N., Chen, Y., Liang, Y., Xiao, M., & Lu, S. (2024) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom52122.2024.10621148 · Cited by 14
“And the rest [21]–[23] adjust analysis configuration adaptively but fail to consider the effect of detection quality on tracking.”
BackgroundJoint Encoding and Enhancement for Low-Light Video Analytics in Mobile Edge Networks
He, Y., Yang, P., Qin, T., Hou, J., & Zhang, N. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3514214 · Cited by 14
“Jiang et al. show that dynamic adjustment of encoding parameters can significantly reduce bandwidth consumption without decreasing the analytics accuracy [49], while Wang et al. show that adaptive control of resolution and frame rate also contributes to transmission efficiency and energy saving [39].”
Edge-Assisted Adaptive Configuration for Serverless-Based Video Analytics
Wang, Z., Zhang, R., Zhang, S., Cheng, W., Wang, W., & Cui, Y. (2025) · IEEE Transactions on Networking · doi:10.1109/ton.2024.3523956 · Cited by 14
“For example, Jiang et al. [5] presented a controller that dynamically picks the optimal video knobs for existing DNN-based video analytics pipelines.”
Influential
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.