一个车辆边缘云框架用于微调深度学习模型的计算分析.
M Jalal Khan1,2, Manzoor Ahmed Khan1,2, Sherzod Turaev1
1College of Information Technology, United Arab Emirates University, Abu Dhabi 15551, United Arab Emirates.
Sensors (Basel, Switzerland)
|April 13, 2024
概括
为自动驾驶汽车部署深度学习模型需要仔细考虑计算层. 这项研究分析了YOLOv8在车辆边缘云架构上的性能,发现了一致的准确性,但根据硬件和环境的推断时间有所不同.
科学领域:
- 计算机科学 计算机科学
- 人工智能的人工智能
- 机器人技术 机器人技术 机器人技术
背景情况:
- 合作,互联和自动化移动 (CCAM) 基础设施对于城市地区的自动驾驶汽车 (AV) 环境感知至关重要.
- 机器学习 (ML) 和深度学习 (DL) 模型的高效部署对于提高CCAM框架内的AV性能至关重要.
- 计算处理层和特定DL模型的选择会影响复杂环境中的AV能力.
研究的目的:
- 为在CCAM基础架构中部署DL模型提出一个计算框架.
- 分析定制训练的YOLOv8 DL模型在车辆边缘云架构中的多种设备和层的有效性.
- 调查DL模型准确性和在分层框架内部署期间的执行时间之间的权衡.
主要方法:
- 开发了一个用于在车辆边缘云层架构中部署DL模型的计算框架.
- 在各种计算层和设备中评估了一种定制训练的YOLOv8 DL模型.
- 在不同的环境条件和硬件配置下测量性能指标,包括精度 (mAP@0.5) 和推断时间.
主要成果:
- 在框架的任何层中,DL模型性能指标 (例如0.842 mAP@0.5) 在不同的设备类型中保持一致.
- 在不同的环境条件下,对象检测任务的推理时间减少.
- 特定的硬件配置显示,推断时间显著减少:与拉斯伯雷Pi (非GPU) 相比,Jetson AGX (非GPU) 减少了72%的时间,与Jetson AGX ARMv8 (非GPU) 相比,Jetson AGX Xavier (GPU) 减少了90%的时间.
结论:
- 计算层和设备的选择在CCAM系统中显著影响DL模型推断时间,而不是它的准确性.
- 研究人员和从业人员可以利用这些发现来优化DL模型部署用于生产AV系统.
- 选择合适的硬件和部署环境是平衡自动驾驶系统性能和效率的关键.
相关概念视频
End Point Prediction: Gran Plot
318
A Gran plot is used to predict the equivalence volume or endpoint of a potentiometric or acid-base titration without reaching the endpoint. Typically, titration data is collected as a function of the titrant's volume up to a point less than the equivalence volume and then transformed into a linear format. The straight line is extended to the x-axis, indicating the necessary titrant volume to achieve the equivalence point.
For potentiometric titration, the Gran plot is created by plotting...
For potentiometric titration, the Gran plot is created by plotting...
318
Differential Leveling
175
Differential leveling is a precise method in surveying used to determine the elevation difference between two points. Its primary goal is to establish accurate vertical measurements to create level surfaces or grade lines critical for designing and constructing infrastructures such as roads, bridges, and buildings.The procedure for differential leveling begins with setting up and leveling the instrument at a point where the benchmark can be seen. The level rod is held on the benchmark (BM), and...
175
Elastic Collisions: Case Study
14.0K
Elastic collision of a system demands conservation of both momentum and kinetic energy. To solve problems involving one-dimensional elastic collisions between two objects, the equations for conservation of momentum and conservation of internal kinetic energy can be used. For the two objects, the sum of momentum before the collision equals the total momentum after the collision. An elastic collision conserves internal kinetic energy, and so the sum of kinetic energies before the collision equals...
14.0K
Manipulation and Analysis
23
GIS manipulation and analysis functions are vital for decision-making and planning. These activities range from data retrieval tasks, such as selecting information based on specific criteria, to advanced analytical techniques that address complex spatial problems.One critical GIS analysis method is overlaying, which combines multiple data layers to examine impacts. For example, overlaying a river-dammed lake boundary with road networks can identify affected infrastructure. Another common...
23


