arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2605.03396cs.AR
收藏

基于FPGA的BNN目标检测设计与实现

Design and Implementation of BNN-Based Object Detection on FPGA

School of Physics, Beihang University(北京航空航天大学物理学院);School of Artificial Intelligence, Beihang University(北京航空航天大学人工智能学院);China Information Technology Designing and Consulting Institute Company Ltd(中国电子信息设计咨询有限公司);Soochow University(苏州大学)

Xuyu Zhao, Yunpeng Wu, Mengyuan Zhu, Haoyu Huang, Xiaoyu Xu, Yanjing Li, Gaolong Zhang, Baochang Zhang

AI总结:

本文在低成本FPGA上实现基于BNN的YOLOv3-tiny目标检测器,采用1位权重和8位激活值,通过固定点卷积提升性能,达到39.6%的mAP50指标。

摘要:

本文实现了一个基于二进制神经网络(BNN)的YOLOv3-tiny-like目标检测器,该检测器在低成本FPGA上运行。网络输入为320*320*3的RGB图像,其主要卷积层使用1位权重和8位激活值,而Conv1和最终检测头使用固定点标准卷积。从训练好的ONNX模型中提取权重、偏置和量化参数,将其转换为定点数,并打包成COE文件存储在Vivado BRAM ROM中。硬件完全用Verilog RTL编写,包括填充、行缓冲、二进制卷积、量化后处理、最大池化和检测头计算。对于Mul_prev由输入通道索引、Div_current由输出通道索引的层,Mul_prev被融合到BNN PE中,以便在累积时应用通道补偿。在PASCAL VOC数据集上,模型达到39.6%的mAP50,计算量为0.098 GFLOPs,参数量为0.74 M。RTL仿真显示,最终的原始检测输出与对应的ONNX节点的相关系数为0.999964,均方误差为0.020027。

英文摘要:

This paper implements a Binary Neural Network (BNN) based YOLOv3-tiny-like object detector on a low-cost FPGA. The network takes 320*320*3 RGB images as input. Its main convolution layers use 1-bit weights and 8-bit activations, while Conv1 and the final detection head use fixed-point standard convolutions. From the trained ONNX model, weights, biases, and quantization parameters are extracted, converted to fixed point, packed into COE files, and stored in Vivado BRAM ROMs. The hardware is written fully in Verilog RTL and includes padding, line buffering, binary convolution, quantization post-processing, max pooling, and detection-head computation. For layers where Mul_prev is indexed by input channel and Div_current by output channel, Mul_prev is fused in-to the BNN PE so that channel-wise compensation is applied during accumulation. On VOC, the model obtains 39.6% mAP50 with 0.098 GFLOPs and 0.74 M parameters. RTL simulation shows that the final raw detection output reaches a correlation coefficient of 0.999964 and a mean absolute error of 0.020027 against the corresponding ONNX node.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑