Optimizing the Objective Detection for RISC-Ⅴ Architecture
Objective detection is one of the most important research directions in the field of computer vision and is widely used in fields such as intelligent surveillance,autonomous driving,and medical image analysis.Faced with a continuous stream of application scenarios,objective detection algorithms often need to be deployed on specific hardware platforms.Optimizing the ob-jective detection algorithm according to the characteristics of the hardware platform can greatly improve the algorithm's inference efficiency.In recent years,RISC-Ⅴ has attracted widespread attention from the academic and industrial communities due to its features of being streamlined,open-source,and customizable.It has developed rapidly and has become the third major CPU ar-chitecture following X86 and ARM.This study focused on the vector extension of RISC-Ⅴ,and optimized the objective detection algorithm through program performance analysis,vectorization,memory access optimization,loop unrolling,and other technolo-gies.It was deployed and tested on simulators and RISC-Ⅴ development boards.The experiments show that,compared with the in-itial version of the algorithm,the optimized version has improved single-threaded inference efficiency by more than 300％.This study verifies the effectiveness of RISC-Ⅴ vector extensions in optimizing object detection algorithms,providing valuable experi-ence and references for future application porting and algorithm optimization on the RISC-Ⅴ platform.
RISC-Ⅴobject detectionvectorsingle instruction multiple data
NETL
NSTL
万方数据
