param is too old, please regenerate!
Closed this issue · 4 comments
First, I use the ncnn models in https://github.com/ouyanghuiyu/chineseocr_lite/tree/onnx/models_ncnn, but I have an error which is "param is too old, please regenerate!"
For solve this problem, I converter a new ncnn model use ncnn project and the onnx models in "https://github.com/DayBreak-u/chineseocr_lite/tree/onnx/models".
But then,I got a new error when I compile the project with the new model. The error is "layer Shape not exists or registered".
Thanks
ncnn 20220420，更新后模型支持有变化，需要重新用新版ncnn自带的onnx2ncnn转换模型
一般转换之前，onnx先做simple简化网络，再进行转换
可以直接使用整合好的工程代码
https://github.com/benjaminwan/OcrLiteAndroidNcnn/releases/tag/1.7.0
Project_OcrLiteAndroidNcnn-1.7.0.7z
https://github.com/benjaminwan/OcrLiteAndroidNcnn/releases/download/1.7.0/Project_OcrLiteAndroidNcnn-1.7.0.7z
ncnn 20220420，更新后模型支持有变化，需要重新用新版ncnn自带的onnx2ncnn转换模型 一般转换之前，onnx先做simple简化网络，再进行转换
可以直接使用整合好的工程代码 https://github.com/benjaminwan/OcrLiteAndroidNcnn/releases/tag/1.7.0 Project_OcrLiteAndroidNcnn-1.7.0.7z https://github.com/benjaminwan/OcrLiteAndroidNcnn/releases/download/1.7.0/Project_OcrLiteAndroidNcnn-1.7.0.7z
wow, Thank you very much！
I have used the latest version of onnx2ncnn to convert the model.But I got the error:"layer Shape not exists or registered"
Is the problem that I do not simplify the onnx net?
If this is the real problem, is there a tutorial to simplify the onnx net for ncnn?
I'll try your release project, but I want to learn how to simplify the onnx net.
I am appreciate of you.
Thank you very much!