一个基于OpenCL的FPGA加速器,用于更快的R-CNNN
Jianjing An1,2, Dezheng Zhang1,2, Ke Xu1,2
1Institute of Information Science, Beijing Jiaotong University, Beijing 100044, China.
Entropy (Basel, Switzerland)
|July 8, 2023
概括
这项研究引入了一种新的FPGA加速器,用于更快的R-CNN对象检测,显著提高性能. 硬件和软件的共同设计实现了比现有的Faster R-CNN加速器提高10倍的吞吐量.
科学领域:
- 计算机工程 计算机工程
- 人工智能的人工智能
- 硬件加速器 硬件加速器
背景情况:
- 基于卷积神经网络 (CNN) 的对象检测算法,特别是Faster R-CNN,由于高计算和内存复杂性而面临挑战.
- 现有的硬件加速器设计主要集中在像YOLO这样的单阶段探测器上,留下了一个有效的更快的R-CNN实现的空白.
- 现场可编程网关数组 (FPGA) 为复杂的深度学习模型的定制硬件加速提供了一个有前途的平台.
研究的目的:
- 提出一个软硬件共同设计方案,用于在使用OpenCL的FPGA上实现更快的R-CNN对象检测算法.
- 设计一个高效的,深层管道的FPGA硬件加速器,能够支持Faster R-CNN的各种骨干网络.
- 通过固定点定量化和层融合等技术优化Faster R-CNN算法用于硬件实现.
主要方法:
- 开发一个深管道FPGA硬件加速器,为快速R-CNN量身定制.
- 实现硬件意识的软件算法,包括固定点定量化,层融合和多批次感兴趣区域 (RoIs) 探测器.
- 利用OpenCL进行软硬件共同设计,并设计一个端到端的空间探索方案,用于性能和资源评估.
主要成果:
- 拟议的FPGA加速器在172 MHz的工作频率下实现了846.9 GOP/s的峰值吞吐量.
- 与最先进的Faster R-CNN加速器相比,演示了10倍的推断吞吐量改进.
- 实现了2.1倍的推断吞吐量改进,而不是一个阶段的YOLO加速器.
结论:
- 拟议的软硬件联合设计方案有效地解决了FPGAs上的Faster R-CNN的计算和内存挑战.
- 开发的FPGA加速器提供了显著的性能提升,使其成为实时对象检测的竞争性解决方案.
- 这项工作推动了复杂的深度学习模型的硬件加速领域的发展,特别是对于双阶段对象检测器.
相关概念视频
Parallel Processing
185
The brain processes sensory information rapidly due to parallel processing, which involves sending data across multiple neural pathways at the same time. This method allows the brain to manage various sensory qualities, such as shapes, colors, movements, and locations, all concurrently. For instance, when observing a forest landscape, the brain simultaneously processes the movement of leaves, the shapes of trees, the depth between them, and the various shades of green. This enables a quick and...
185
Acceleration Vectors
8.1K
In everyday conversation, accelerating means speeding up. Acceleration is a vector in the same direction as the change in velocity, Δv, therefore the greater the acceleration, the greater the change in velocity over a given time. Since velocity is a vector, it can change in magnitude, direction, or both. Thus acceleration is a change in speed or direction, or both. For example, if a runner traveling at 10 km/h due east slows to a stop, reverses direction, and continues their run at 10 km/h...
8.1K
Accelerators
95
Accelerators in concrete serve as admixtures to speed up the hardening process, enabling the concrete to achieve early strength faster. Although accelerators do not necessarily impact the time it takes concrete to set, they reduce this time in practice. A common accelerator is calcium chloride, which is particularly useful for hastening early strength development in cold weather or for rapid repair jobs that require quick heat generation after mixing.
The effectiveness of calcium chloride can...
The effectiveness of calcium chloride can...
95
Clamper Circuit
492
A clamper circuit, also known as a DC restorer, represents a specialized variant of the rectifier circuit, notable for its method of taking the output across the diode rather than the capacitor. This configuration lends to several distinctive applications, particularly in handling square wave inputs.
Within this circuit, the diode's orientation prompts the capacitor to charge up to the level of the most negative peak of the input signal. Upon reaching this state, the diode ceases to...
Within this circuit, the diode's orientation prompts the capacitor to charge up to the level of the most negative peak of the input signal. Upon reaching this state, the diode ceases to...
492
Average Acceleration
9.7K
The importance of understanding acceleration spans our day-to-day experiences, as well as the vast reaches of outer space and the tiny world of subatomic physics. In everyday conversation, to accelerate means to speed up. For instance, we are familiar with the acceleration of our car; the harder we apply our foot to the gas pedal, the faster we accelerate. The greater the acceleration, the greater the change in velocity over a given time. Acceleration is widely seen in experimental physics. In...
9.7K
Improving Translational Accuracy
11.7K
Base complementarity between the three base pairs of mRNA codon and the tRNA anticodon is not a failsafe mechanism. Inaccuracies can range from a single mismatch to no correct base pairing at all. The free energy difference between the correct and nearly correct base pairs can be as small as 3 kcal/ mol. With complementarity being the only proofreading step, the estimated error frequency would be one wrong amino acid in every 100 amino acids incorporated. However, error frequencies observed in...
11.7K


