A Real-Time Object Detection Processor With xnor-Based Variable-Precision Computing Unit

Object detection algorithms using convolutional neural networks (CNNs) achieve high accuracy, but it is challenging to realize real-time object due their computational complexity, especially on resource-constrained mobile platforms. In this article, we propose an algorithm-hardware co-optimization approach designing a system. We first develop compact model based binarized network (BNN), which employs new layer structure, the DenseToRes layer, mitigate information loss deep quantization. also efficient processor that runs with throughput limited hardware rescources. resource-efficient processing unit supporting variable precision minimal overheads. Implemented in field-programmable gate array (FPGA), achieves 64.51 frames/s 64.92 mean average (mAP) accuracy. Compared prior FPGA-based designs for detection, our design competitive accuracy and lower implementation costs.