search

RangiLyu / nanodet

NanoDet-Plusāš”Super fast and lightweight anchor-free object detection model. šŸ”„Only 980 KB(int8) / 1.8MB (fp16) and run 97FPS on cellphonešŸ”„
Apache License 2.0
5.71k stars 1.04k forks source link

converting issue from .pth to onnx then to ncnn #556

Closed gedance closed 7 months ago

gedance commented 7 months ago

Hello,

I'm facing some issues with converting properly from onnx to ncnn. For some reasons, ncnn param looks wierd (added some stuff like input.4.. onnx:: etc.)

image

Does anyone know how to solve this problem to get correctly converted ncnn files?

I'm using below versions in Anaconda environment:

Python 3.8.18 (default, Sep 11 2023, 13:40:15) [GCC 11.2.0] :: Anaconda, Inc. on linux Type "help", "copyright", "credits" or "license" for more information.

import torch torch.version '1.12.0+cu102' import onnx onnx.version '1.12.0'

Thanks!

gedance commented 7 months ago

When I used yolox to ncnn, there was no things like "onnx::" in the param file, so I thought current .param above is incorrect. But the above .param and .bin ncnn files work fine when inference from c++ code. So closing the issue.