Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

关于MNN1.1.0版本问题 #92

Closed
tunmx opened this issue Dec 21, 2020 · 10 comments
Closed

关于MNN1.1.0版本问题 #92

tunmx opened this issue Dec 21, 2020 · 10 comments

Comments

@tunmx
Copy link

tunmx commented Dec 21, 2020

请问是否有检查过在mnn 1.1.0版本下存在的问题,我在1.1.0版本下使用仓库内的mnn模型推理出来的结果会有问题,然后用1.1.0的转换器通过onnx转出的mnn模型也会有一样的问题

@RangiLyu
Copy link
Owner

@xieruicheng

@ruicx
Copy link
Contributor

ruicx commented Dec 22, 2020

请使用 MNN 1.0.0MNN 1.1.0 下很多从ONNX转的检测模型的结果都对不上

@ruicx
Copy link
Contributor

ruicx commented Jan 12, 2021

C++ 测试了刚刚发布的 MNN 1.1.2 版本, 检测结果看起来基本正常,但是目标的得分和Pytorch还是有区别。MNN 1.0.0 的得分和Pytorch小数点后四位是完全一致的。

Python 版本在pip任然是 Dec 22, 2020 的 release,此前测试有问题,这次没有测试。

@eeric
Copy link

eeric commented Jan 14, 2021

test myself,
nanodet model convert to MNN one by 1.0.0, not use 1.1.0 or 1.1.2
notice: model converted by MNN-1.0.0 that can be use to vs2017 demo_mnn with MNN-1.1.2 library

@scutzhe
Copy link

scutzhe commented Jan 18, 2021

mnn version 1.0.0 is OK、

@job2003
Copy link

job2003 commented Jan 30, 2021

请使用 MNN 1.0.0MNN 1.1.0 下很多从ONNX转的检测模型的结果都对不上

如何安装python版本的1.0.0,我使用
pip install MNN默认安装1.1.0版本。
如果使用pip install MNN==1.0.0会出现错误。
ERROR: Could not find a version that satisfies the requirement MNN==1.0.0 (from versions: 1.0.5, 1.0.6, 1.1.0)
ERROR: No matching distribution found for MNN==1.0.0

@ruicx
Copy link
Contributor

ruicx commented Jan 31, 2021

如何安装python版本的1.0.0,我使用
pip install MNN默认安装1.1.0版本。
如果使用pip install MNN==1.0.0会出现错误。
ERROR: Could not find a version that satisfies the requirement MNN==1.0.0 (from versions: 1.0.5, 1.0.6, 1.1.0) ERROR: No matching distribution found for MNN==1.0.0

目前 pypi 上只有py27,py35-py37的版本,其他python版本需要自己编译。

另,刚刚测试了 MNN 1.0.6,检测结果正确

@Co2Link
Copy link

Co2Link commented Feb 14, 2021

利用仓库里提供的nanodet-320.mnn., MNN 1.1.3 的结果与pytorch小数点后4位完全一致

@scutzhe
Copy link

scutzhe commented Feb 24, 2021

mnn version 1.0.0 is OK、

but backbone is ghostNet,mnn version 1.0.0 error, mnn version 1.1.0 error too

@scutzhe
Copy link

scutzhe commented Feb 24, 2021

mnn version 1.0.0 is OK、

but backbone is ghostNet,mnn version 1.0.0 error, mnn version 1.1.0 error too

open /tools/converter/source/onnx/Clip.cpp and comment out two lines with DCHECK_EQ.
Then rebuild and it works well

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

7 participants