Live Video Analytics at Scale with Approximation and Delay-Tolerance

By Haoyu Zhang, Ganesh Ananthanarayanan, Peter Bodík, Matthai Philipose, Paramvir Victor Bahl, Michael J. Freedman. Symposium on Networked Systems Design and Implementation, 2017

485 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/4aa43a0a70a39efdc37fc7218747a9806fe8eeb6

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Sergey Levine
2. Dawn Song
3. Pieter Abbeel
4. Ion Stoica
5. Yoshua Bengio
6. Jiasi Chen
7. Percy Liang
8. Eric P. Xing

Citations in the corpus

• Multi-tenant GPU Clusters for Deep Learning Workloads: Analysis and Implications (2018, methodology): “More recently, approximation is used in batch processing [7, 9, 49] and live video analytics [56], as well as in designing system components such as DRAM [42].”
• A Berkeley View of Systems Challenges for AI (2017, background): “mising accuracy, and it has been already employed in recent video recognition systems [59, 115].”
• NestDNN: Resource-Aware Multi-Tenant On-Device Deep Learning for Continuous Mobile Vision (2018, methodology): “Taking a traffic surveillance drone as an example: an application that counts vehicles to detect traffic jams does not need high accuracy but requires low latency; an application that reads license plates needs high plate reading accuracy but does not require real-time response [39].”
• Computation Offloading Toward Edge Computing (2019, methodology): “Evaluation results show that VideoEdge outperforms the two-tier (cameras and the cloud) video analytics system— VideoStorm [162].”
• InferLine: latency-aware provisioning and scaling for prediction serving pipelines (2020, methodology): “VideoStorm [40] explores the design of a streaming video processing system with a distributed design with pipeline operators provisioned across compute nodes and explores the combinatorial search space of hardware and model configurations.”
• DeepDecision: A Mobile Deep Learning Framework for Edge Video Analytics (2018, background): “[26] performs resource profiling similar to our work, but focuses on server-side scheduling whereas we focus on client-side decisions.”
• Deep Learning With Edge Computing: A Review (2019, background): “VideoStorm [93] was one of the first works in this space, and profiles these tradeoffs to choose the right DNN configuration for each request, to meet the accuracy and latency goals.”
• Edge Assisted Real-time Object Detection for Mobile Augmented Reality (2019, background): “VideoStorm [45] and Chameleon [46] achieve higher accuracy video analytics with the same amount of computational resources on the cloud by adapting the video configurations.”
• Focus: Querying Large Video Datasets with Low Latency and Low Cost (2018, background): “, [96]) mainly focus on general stream processing challenges such as load shedding, fault tolerance, distributed execution, or limited network bandwidth.”
• Reducto: On-Camera Filtering for Resource-Efficient Real-Time Video Analytics (2020, background): “Systems such as Chameleon [41] and VideoStorm [80] reduce backend computation costs by profiling different configurations of pipeline knobs (e.”
• Chameleon: scalable adaptation of video analytics (2018, methodology): “We build on prior work [32] where video modules constantly “listen” and prepare for any configuration updates.”
• Real-Time Video Analytics: The Killer App for Edge Computing (2017, background): “, the time between arrival and processing of the frames [4].”