FPGA-based Acceleration of Neural Network for Image Classification using Vitis AI
This work addresses efficiency issues for computer vision applications, but it is incremental as it applies an existing method (Vitis AI) to a specific hardware setup.
The paper tackled the problem of insufficient throughput and high power consumption in CNN-based image classification by accelerating a CNN on an FPGA using Vitis AI, achieving 3.33-5.82x higher throughput and 3.39-6.30x higher energy efficiency compared to CPU and GPU baselines on the CIFAR-10 dataset.
In recent years, Convolutional Neural Networks (CNNs) have been widely adopted in computer vision. Complex CNN architecture running on CPU or GPU has either insufficient throughput or prohibitive power consumption. Hence, there is a need to have dedicated hardware to accelerate the computation workload to solve these limitations. In this paper, we accelerate a CNN for image classification with the CIFAR-10 dataset using Vitis-AI on Xilinx Zynq UltraScale+ MPSoC ZCU104 FPGA evaluation board. The work achieves 3.33-5.82x higher throughput and 3.39-6.30x higher energy efficiency than CPU and GPU baselines. It shows the potential to extract 2D features for downstream tasks, such as depth estimation and 3D reconstruction.