onnx如何转ncnn

yolov5的onnx怎么转成ncnn啊，为什么我转出来没有focus层

用netron工具打开param，找到对应focus的部分，修改，如这里所示 https://blog.csdn.net/qq_45057749/article/details/115013509

方案来自 梦想橡皮擦 狂飙组基于 GPT 编写的 “程秘”

转换 YOLOv5 模型到 ncnn 格式可以使用 OpenVINO 转换工具或者其他类似的工具。如果转换后没有 Focus 层，可能是因为 Focus 层不是所有框架都支持的

• 这篇文章：pytorch onnx转ncnn 也许有你想要的答案，你可以看看
• 除此之外, 这篇博客: yolov5 ncnn 模型 和 Onnx模型精度校验 |【output层对比】中的 yolov5s_5_opt.param【ncnn】 推理输出 部分也许能够解决你的问题, 你可以仔细阅读以下内容或跳转源博客中阅读:
  

• python yolov5sInfer_ncnn_out.py dog.png 
  
  self.net.output_names() ['output', '417', '437']
  out_names[0]: -- output 
  type(ncnn_out) <class 'numpy.ndarray'>
  
  ncnn_out.shape  (3, 6400, 85)
  printNumpy()
  -0.138561487197875977  -0.009932726621627808  -0.258105278015136719  -0.380718827247619629  -12.517526626586914062  -0.669204592704772949  -4.980797290802001953  -3.500458240509033203  -6.847023487091064453  -5.911782741546630859  
  -6.206901073455810547  -6.214505195617675781  -5.343970298767089844  -4.193430900573730469  -4.471679210662841797  -5.453211784362792969  -4.596863269805908203  -5.232987403869628906  -4.264289379119873047  -3.769028663635253906  
  
  
  out_names[1]: -- 417 
  type(ncnn_out) <class 'numpy.ndarray'>
  ncnn_out.shape  (3, 1600, 85)
  printNumpy()
  -0.974951028823852539  -0.091828197240829468  -0.078549027442932129  -0.824775218963623047  -10.806281089782714844  -1.558434247970581055  -5.618755340576171875  -4.445729255676269531  -5.951706886291503906  -6.004321575164794922  
  -6.262535095214843750  -6.369493007659912109  -5.844290256500244141  -5.516223430633544922  -4.714602470397949219  -6.453022003173828125  -5.975280284881591797  -6.670634269714355469  -4.930451393127441406  -5.172771453857421875  
  
  
  out_names[2]: -- 437 
  type(ncnn_out) <class 'numpy.ndarray'>
  ncnn_out.shape  (3, 400, 85)
  printNumpy()
  -0.341668695211410522  0.147016882896423340  -0.491063684225082397  -0.354453921318054199  -9.955618858337402344  -1.710951924324035645  -5.311002731323242188  -4.368266105651855469  -5.934424400329589844  -5.750700473785400391  
  -5.760524749755859375  -5.578440666198730469  -5.427252769470214844  -5.289523124694824219  -5.797097206115722656  -6.159203052520751953  -5.641136169433593750  -6.624048709869384766  -4.286302089691162109  -5.322281360626220703  
  
  
  16 = 0.79546 at 136.28 283.78 177.64 x 440.98
  
  7 = 0.63773 at 467.44 101.13 222.54 x 127.14
  
  2 = 0.45113 at 466.72 99.08 221.51 x 128.81
  
  1 = 0.29964 at 131.10 162.85 453.83 x 552.17
  
  

  ---