2022
DOI: 10.1587/elex.19.20220067
|View full text |Cite
|
Sign up to set email alerts
|

HFOD: A hardware-friendly quantization method for object detection on embedded FPGAs

Abstract: There are two research hotspots for improving performance and energy efficiency of the inference phase of Convolutional neural networks (CNNs). The first one is model compression techniques while the second is hardware accelerator implementation. To overcome the incompatibility of algorithm optimization and hardware design, this paper proposes HFOD, a hardware-friendly quantization method for object detection on embedded FPGAs. We adopt a channel-wise, uniform quantization method to compress YOLOv3-Tiny model.… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
3

Citation Types

0
3
0

Year Published

2022
2022
2024
2024

Publication Types

Select...
4

Relationship

0
4

Authors

Journals

citations
Cited by 4 publications
(3 citation statements)
references
References 23 publications
0
3
0
Order By: Relevance
“…Many accelerators with execution prediction are also proposed to reduce the computation of DNNs [8,9]. Quantization is a common step in DNNs deployment [10,11,12]. However, dynamic bit-width quantization has not been well studied.…”
Section: Introductionmentioning
confidence: 99%
“…Many accelerators with execution prediction are also proposed to reduce the computation of DNNs [8,9]. Quantization is a common step in DNNs deployment [10,11,12]. However, dynamic bit-width quantization has not been well studied.…”
Section: Introductionmentioning
confidence: 99%
“…The designs proposed in [16,17,21] can deal with convolutions of several common kernel sizes, but it is still not applicable to convolutions of any kernel sizes. The authors in [22][23][24][25][26] adopt multiple computing engines to deal with the convolution with different kernel sizes for improving performance. However, this design costs too much hardware resources, which is not suitable for low-cost FPGAs.…”
Section: Introductionmentioning
confidence: 99%
“…Thus, without elaborately design, this mismatch between data and computation elements may lead to ultra-low utilization of FPGA resources, which is undesirable for low-cost FPGAs. For example, in [16][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31], although the size of the CNN model is significanly reduced, the bit utilization of FPGA resources is very low. To solve the above problems, this brief proposes an efficient hardware accelerator for low-bit quantized lightweight CNN models and the contribution can be summarized as follows:…”
Section: Introductionmentioning
confidence: 99%