Convolutional neural network implementations using Vitis AI

Recently, Xilinx has provided a field programmable gate array (FPGA)-based Vitis AI development environment, which is a deep learning framework to accel-erate AI operations and to seek a suitable neural network construction for a target application. We have implemented convolutional neural networks of three types onto the Vitis AI development environment and then we have evaluated their performance, power consumption, design man-hours, and so on. Results confirmed the Vitis AI benefits. Most notably, the FPGA platform power consumption is 4.96 times less than that of a GPU.