jetson nano 部署使用tensorRT加速YOLOv4tiny

     我在nano上测试了darknet的tolov4和yolov4 tiny两种权重，前者帧数只有一点几帧，后者能在10帧上下，英伟达平台支持tensorRT来优化权重实现加速，优化后有20多帧，但是也会降低识别准确度，这个看个人需求吧。
     参考了下面两个大佬的两篇博客，pycuda跟我版本不同的可以看下下面文章的配置步骤
    
     [pycuda安装]
     (https://blog.csdn.net/weixin_44501699/article/details/106470671?ops_request_misc=%257B%2522request%255Fid%2522%253A%2522168445759716782427492812%2522%252C%2522scm%2522%253A%252220140713.130102334..%2522%257D&request_id=168445759716782427492812&biz_id=0&utm_medium=distribute.pc_search_result.none-task-blog-2~all~sobaiduend~default-1-106470671-null-null.142%5Ev87%5Econtrol,239%5Ev2%5Einsert_chatgpt&utm_term=jetson%20nano%20%E5%AE%89%E8%A3%85pycuda&spm=1018.2226.3001.4187)
     
     [配置过程参考]
     (https://blog.csdn.net/qq_51963216/article/details/121032566?ops_request_misc=&request_id=&biz_id=102&utm_term=jetson%20nano%E4%BD%BF%E7%94%A8yolo%E5%B8%A7%E7%8E%87%E4%BD%8E&utm_medium=distribute.pc_search_result.none-task-blog-2~all~sobaiduweb~default-0-1