Performance and Memory Trade-offs of Deep Learning Object Detection in Fast Streaming High-Definition Images

Srivastava, Aishwarya; Nguyen, Dung; Aggarwal, Siddhant; Luckow, André; Duffy, Edward B.; Kennedy, Ken; Ziolkowski, Marcin; Apon, Amy

doi:10.1109/bigdata.2018.8622389

Cited by 9 publications

(14 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The results shown are representative of the range of results and are the most likely combinations of models and architecture to be selected for our application. Results for hardware platforms P100, V100 PCIe, and V100 SXM2 are reported in our previous work . Larger batch sizes require larger memory to store input tensors, intermediate representations, and output tensors during computations.…”

Section: Performance and Memory Tradeoffs Of Object Detection Modelsmentioning

confidence: 92%

“…This task is referred to as object detection. Detailed reviews of object detection methods have been provided elsewhere in the literature (eg, see Reference ). Here we give a summary of the primary considerations in selecting a particular object detection model.…”

Section: Background and Literature Reviewmentioning

confidence: 99%

“…In this article, we summarize results on two popular GPU devices that are available on AWS. Evaluation across a larger range of hardware is performed in Reference . To determine the inference time of various object detection models, we run experiments on the Clemson University supercomputer, known as Palmetto.…”

Section: Performance and Memory Tradeoffs Of Object Detection Modelsmentioning

confidence: 99%

“…The complete visual view of a car in the synthetic workload is represented by 95 images. The right side is zoomed in on four of the tiles and shows the objects detected [Color figure can be viewed at wileyonlinelibrary.com]…”

Section: Performance and Memory Tradeoffs Of Object Detection Modelsmentioning

confidence: 99%

“…This figure shows the time for one image only of the 95 images required to cover the entire car. The left figure shows results for V100 PCIe, and the right figure shows results for V100 SXM2 . The labels used in this figure are shown in Figure 5.…”

Section: Performance and Memory Tradeoffs Of Object Detection Modelsmentioning

confidence: 99%

See 4 more Smart Citations

Deployment of a cloud pipeline for real‐time visual inspection using fast streaming high‐definition images

et al. 2020

Self Cite

View full text Add to dashboard Cite

We investigate the challenges of building an end-to-end cloud pipeline for real-time intelligent visual inspection system for use in automotive manufacturing. Current methods of visual detection in automotive assembly are highly labor intensive, and thus prone to errors. An automated process is sought that can operate within the real-time constraints of the assembly line and can reduce errors. Components of the cloud pipeline include capture of a large set of high-definition images from a camera setup at the assembly location, transfer and storage of the images as needed, execution of object detection, and notification to a human operator when a fault is detected. The end-to-end execution must complete within a fixed time frame before the next car arrives in the assembly line. In this article, we report the design, development, and experimental evaluation of the tradeoffs of performance, accuracy, and scalability for a cloud system. K E Y W O R D Scloud, end-to-end pipeline, latency, real-time system, visual inspection INTRODUCTIONMachine learning has proven to be highly effective in providing solutions to wide variety of challenging problems in several domains. In the automotive domain, machine learning and deep learning have several potential applications both inside and outside the vehicle. 1 Advanced driving assistance systems (ADAS) and system for self-driving cars operate within the vehicle while systems outside the vehicle indulge in the manufacturing assembly plant systems. 2 Building an end-to-end system pipeline that facilitates continuous delivery requires careful consideration of various tradeoffs. The pipeline should be responsive, scalable for different workloads, fault tolerant, replicable, and cost-effective. Much research lately focuses on optimization of machine learning algorithms, developing frameworks for these algorithms, analyzing the tradeoffs of various parameters associated with these models, or developing a pipeline or system architecture for automatically deploying the infrastructure. The last two topics are main focus of this research. Our work focuses on the end-to-end system architecture for visual inspection using an Amazon Web Services (AWS) implementation. We study various tradeoffs associated with the object detection models using high-definition images. We also discuss interesting use cases for implementation tradeoffs when considering the latency and cost for the end-to-end visual inspection pipeline. 868Visual inspection in the automotive manufacturing is a highly labor intensive and slow process. The quality checks in automotive manufacturing plant involve high-quality visual inspection to ensure that the parts are in the correct locations, have the right shape, that the product does not have any missing parts, and that the automobile is free from scratches, dents, or blemishes. Visual inspection can be facilitated through the use of deep learning models that are deployed on the cloud. Models that are connected to a set of high-definition cameras located at the assembly stat...

show abstract

Section: Performance and Memory Tradeoffs Of Object Detection Modelsmentioning

confidence: 92%

Section: Background and Literature Reviewmentioning

confidence: 99%

Section: Performance and Memory Tradeoffs Of Object Detection Modelsmentioning

confidence: 99%

Section: Performance and Memory Tradeoffs Of Object Detection Modelsmentioning

confidence: 99%

Section: Performance and Memory Tradeoffs Of Object Detection Modelsmentioning

confidence: 99%

See 3 more Smart Citations

Deployment of a cloud pipeline for real‐time visual inspection using fast streaming high‐definition images

et al. 2020

Self Cite

View full text Add to dashboard Cite

show abstract

Künstliche Intelligenz: Anwendungen und Werkzeuge in der Automobilindustrie

Luckow,

Eirich,

Demtröder

2023

Künstliche Intelligenz Und Data Science in Theorie Und Praxis

View full text Add to dashboard Cite

A road map for planning-deploying machine vision artifacts in the context of industry 4.0

Silva

Canciglieri

Rudek

2021

Journal of Industrial and Production Engineering

View full text Add to dashboard Cite

Performance and Memory Trade-offs of Deep Learning Object Detection in Fast Streaming High-Definition Images

Cited by 9 publications

References 8 publications

Deployment of a cloud pipeline for real‐time visual inspection using fast streaming high‐definition images

Deployment of a cloud pipeline for real‐time visual inspection using fast streaming high‐definition images

Künstliche Intelligenz: Anwendungen und Werkzeuge in der Automobilindustrie

A road map for planning-deploying machine vision artifacts in the context of industry 4.0

Contact Info

Product

Resources

About