YOLOv3在FPGA开发板上的部署

his article studies the deployment of the YOLOv3 object detection algorithm on FPGA hardware, using the AXU2CGB-E development board (equipped with Xilinx Zynq UltraScale+ MPSoC chip). The Vitis AI toolchain is utilized to achieve accelerated inference for the YOLOv3 model, and real-time video streams are obtained through a USB camera for object detection. The article first briefly introduces the principles and model structure of the YOLOv3 algorithm; then it explains the relevant work we did before starting the experiments; next, it designs and implements a system deployment scheme based on the AXU2CGB-E development board, including model quantization, compilation, and Python inference scripts; finally, the deployment effects are demonstrated through experiments, including detection results images and frame rate performance. Experimental results show that using FPGA accelerators can effectively achieve real-time object detection capabilities with YOLOv3.