Lightweight real-time stereo matching algorithm for AI chips

doi:10.1016/j.comcom.2022.06.018

	Lightweight real-time stereo matching algorithm for AI chips
	Liu, Yi 1; Wang, Wenhao 2,3; Xu, Xintao 1; Guo, Xiaozhou 1,3; Gong, Guoliang 1,3; Lu, Huaxiang 1,3,4,5
	2023-02-01
发表期刊	COMPUTER COMMUNICATIONS
ISSN	0140-3664
卷号	199 页码:210-217
摘要	AI chips have developed rapidly and achieved remarkable acceleration effects in the corresponding algorithm field in recent years. However, deep learning algorithms are changing rapidly, including many operators that AI chips and inference frameworks cannot use in the short term. To solve the problem that it is challenging to deploy a stereo matching algorithm based on binocular vision on AI chips, this paper proposes a multi-stage unsupervised lightweight real-time depth estimation algorithm for AI chips called TradNet. TradNet combines the traditional matching algorithm with a convolutional neural network and uses convolution directly supported by AI chips to realize the structure of the traditional matching algorithm. TradNet is composed of operators directly supported by current AI chips, which reduces the computational complexity of the algorithm, and greatly improves the compatibility of the stereo matching algorithm with existing AI chips. Compared with the deep learning-based multi-stage binocular disparity algorithm AnyNet, the accuracy is improved by 5.12%, and the inference speed is only 12.7%. Compared with the matching-based binocular disparity algorithm BM, the accuracy is improved by 25.24%, and the inference speed is only 48.7%. Our final model can process 1280x720 resolution images within a range of 60-80 FPS on an NVIDIA TITAN Xp. It achieves 28FPS on a 1TOPS (Tera Operations Per Second) custom AI chip, and the power consumption is 0.88 W.
关键词	AI chips Stereo matching Lightweight network Unsupervised learning Multi-stage stereo matching
DOI	10.1016/j.comcom.2022.06.018
收录类别	SCI
语种	英语
资助项目	Beijing Academy of Artificial Intelligence (BAAI) ; CAS Strategic Leading Science and Technology Project[XDA27040303] ; CAS Strategic Leading Science and Technology Project[XDA18040400] ; CAS Strategic Leading Science and Technology Project[XDB44000000] ; National Natural Science Foundation of China[U19A 2080] ; National Natural Science Foundation of China[U1936106] ; High Technology Project[31513070501] ; High Technology Project[1916312ZD0090-2201]
WOS研究方向	Computer Science ; Engineering ; Telecommunications
WOS类目	Computer Science, Information Systems ; Engineering, Electrical & Electronic ; Telecommunications
WOS记录号	WOS:000916921000001
出版者	ELSEVIER
引用统计
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/20011
专题	中国科学院计算技术研究所期刊论文
通讯作者	Gong, Guoliang
作者单位	1.Chinese Acad Sci, Inst Semicond, Beijing, Peoples R China 2.Chinese Acad Sci, Inst Comp Technol, Beijing, Peoples R China 3.Univ Chinese Acad Sci, Beijing, Peoples R China 4.Chinese Acad Sci, Ctr Excellence Brain Sci & Intelligence Technol, Beijing, Peoples R China 5.Beijing Key Lab Semicond Neural Network Intellige, Beijing, Peoples R China 6.Univ Sci & Technol China, Hefei, Peoples R China
推荐引用方式 GB/T 7714	Liu, Yi,Wang, Wenhao,Xu, Xintao,et al. Lightweight real-time stereo matching algorithm for AI chips[J]. COMPUTER COMMUNICATIONS,2023,199:210-217.
APA	Liu, Yi,Wang, Wenhao,Xu, Xintao,Guo, Xiaozhou,Gong, Guoliang,&Lu, Huaxiang.(2023).Lightweight real-time stereo matching algorithm for AI chips.COMPUTER COMMUNICATIONS,199,210-217.
MLA	Liu, Yi,et al."Lightweight real-time stereo matching algorithm for AI chips".COMPUTER COMMUNICATIONS 199(2023):210-217.