Magic-pipe: Self-optimizing video analytics pipelines

G Coviello, Y Yang, K Rao, S Chakradhar - Proceedings of the 22nd …, 2021 - dl.acm.org
… latency, accuracy and scalability of video analytic pipelines. … the feature extractor while
Chameleon only focuses on object … is not in the scope of Chameleon. Furthermore, as shown in …

Batch adaptative streaming for video analytics

L Zhang, Y Zhang, X Wu, F Wang, L Cui… - … -IEEE Conference on …, 2022 - ieeexplore.ieee.org
Chameleon [19] adopts a video streaming adaptation by periodically profiling the beginning
part of each batch to output the configuration update. AWStream [20] uses a streaming …

Lightweight online profiling-based configuration adaptation for video analytics system in edge computing

WJ Kim, CH Youn - IEEE Access, 2020 - ieeexplore.ieee.org
… The profiling cost of Chameleon is even much higher than its … inference cost of Chameleon
for video analytics on VIRAT … Chameleon: scalable adaptation of video analytics." …

Scaling video analytics on constrained edge nodes

C Canel, T Kim, G Zhou, C Li, H Lim… - Proceedings of …, 2019 - proceedings.mlsys.org
… This section provides an overview of video analytics before delving into the key
challenges … Scaling real-time, wide-area video analytics poses a challenge for bandwidth-limited, …

Soudain: Online adaptive profile configuration for real-time video analytics

K Wu, Y Jin, W Miao, Z Zeng, Z Qian… - 2021 IEEE/ACM 29th …, 2021 - ieeexplore.ieee.org
Chameleon exceed slot frequently, which means ChameleonScalable object detection using
deep neural networks. In … based configuration adaptation for video analytics system in edge …

A Low-Cost Video Analytics System with Velocity Based Configuration Adaptation in Edge Computing

WJ Kim, CH Youn - Advances in Artificial Intelligence and Applied …, 2021 - Springer
… a low-cost video analytics system analyzing multiple video streams … Compared to Chameleon,
the proposed algorithm finds … Meanwhile, Chameleon, which profiles limited configuration …

Scaling video analytics systems to large camera deployments

S Jain, G Ananthanarayanan, J Jiang, Y Shu… - Proceedings of the 20th …, 2019 - dl.acm.org
… The two approaches represent a tradeoff— the data-driven approach can better adapt to …
Chameleon [18] exploits content similarity across cameras to amortize query profiling costs, …

{RECL}: Responsive {Resource-Efficient} continuous learning for video analytics

M Khani, G Ananthanarayanan, K Hsieh… - … USENIX Symposium on …, 2023 - usenix.org
… Continuous learning has recently shown promising results for video analytics by adapting a
… fundamental limitations so that video analytics are scalable with more consistent accuracy. …

Distream: scaling live video analytics with workload-adaptive distributed edge intelligence

X Zeng, B Fang, H Shen, M Zhang - Proceedings of the 18th Conference …, 2020 - dl.acm.org
… live video analytics system based on the smart camera-edge cluster architecture, that is able
to adapt … systems such as VideoStorm [39], Chameleon [22] and NoScope [25] where video

Ekya: Continuous learning of video analytics models on edge compute servers

R Bhardwaj, Z Xia, G Ananthanarayanan… - … USENIX Symposium on …, 2022 - usenix.org
… 3) We design a scalable resource scheduler for joint … new classes, as well as adapt to the
changing characteristics of the … Chameleon exploits correlations in camera content to amortize …