how to convert yolov5-v6.0 to ncnn?

[nqthai309]

i test with ncnn model yolov5 v60 pre-training and it works fine, but when i manually convert pre-training model yolov5m6.pt -> onnx -> ncnn get error like this. Can someone help me.

=> convert onnx -> ncnn

=> inference ncnn model

[Whu-wxy]

My command:

python export.py --include onnx --weight path/to/x.pt --img 640 --train --simplify

Have you simplified it？

[nqthai309]

I do it in this order, help me !!

yolov5m6.pt ---> onnx simplify

onnx ----> ncnn

run ncnn model

[Whu-wxy]

Have you tried yolov5m/s/n or online conversion ?

[nqthai309]

Have you tried yolov5m/s/n or online conversion ?

I try convert online

download model then optimize file param and bin

build and run yolov5 ncnn

Segmentation fault :(

[yu937861]

I encountered the same problem and tried many times without solving it。

[yu937861]

i test with ncnn model yolov5 v60 pre-training and it works fine, but when i manually convert pre-training model yolov5m6.pt -> onnx -> ncnn get error like this. Can someone help me.

=> convert onnx -> ncnn

=> inference ncnn model

If you solve it, please tell me your solution.

[wangchunin]

python export.py --weights ./yolov5m.pt --include torchscript onnx --train add "--train"

[rgkannan676]

Hi @nihui , Can you please advise on this issue. Thanks.

[rgkannan676]

"python export.py --include onnx --weight path/to/x.pt --img 640 --train --simplify" resolved my error during ncnn conversion.

[nqthai309]

i try convert pretrained model yolov5m6.pt -> onnx with "python3 export.py --include onnx --weights yolov5m6.pt --train --simplify" then convert onnx model -> ncnn model, but when i load them i get this error

"python export.py --include onnx --weight path/to/x.pt --img 640 --train --simplify" resolved my error during ncnn conversion.

[rgkannan676]

You need to change the extract layers in the c++ file. Layers of the current model is mentioned in the error message under “Try”.

[rgkannan676]

The example script is for yolov5s6 model. I tried with yolov5m6 model and didn’t get good results as you can see that there are 4 output layers. Hi @nihui can you please advise what changes are required for yolov5m6 model? Thanks

[nihui]

针对onnx模型转换的各种问题，推荐使用最新的pnnx工具转换到ncnn In view of various problems in onnx model conversion, it is recommended to use the latest pnnx tool to convert your model to ncnn

pip install pnnx
pnnx model.onnx inputshape=[1,3,224,224]

详细参考文档 Detailed reference documentation https://github.com/pnnx/pnnx https://github.com/Tencent/ncnn/wiki/use-ncnn-with-pytorch-or-onnx#how-to-use-pnnx