Xiangrong Chen
Xiangrong Chen
> This is way i convert onxx file on google colab ? At what step do i need to edit or add ? I saw some guide to convert using...
This is converted from your onnx file, have a try, friend param：[https://drive.google.com/file/d/1pzJi98gefN5JiRtppDGm0gW2YGFBkDkY/view?usp=sharing](https://drive.google.com/file/d/1pzJi98gefN5JiRtppDGm0gW2YGFBkDkY/view?usp=sharing) bin：[https://drive.google.com/file/d/1pzJi98gefN5JiRtppDGm0gW2YGFBkDkY/view?usp=sharing](https://drive.google.com/file/d/1pzJi98gefN5JiRtppDGm0gW2YGFBkDkY/view?usp=sharing) I think there is no trouble with your param files and mine, one is fp32 and the...
> > This is converted from your onnx file, have a try, friend param：https://drive.google.com/file/d/1pzJi98gefN5JiRtppDGm0gW2YGFBkDkY/view?usp=sharing bin：https://drive.google.com/file/d/1pzJi98gefN5JiRtppDGm0gW2YGFBkDkY/view?usp=sharing I think there is no trouble with your param files and mine, one is fp32...
Yeah, I am. There is already a pretrained model, so finetune doesn‘t take up too much time
你可以做下实验，我个人认为head部分不需要太重，比较提取特征的大部分工作在骨干进行
版本的问题，可能是你某个库升级导致的情况
这个应该可以:https://github.com/ppogg/shufflev2-yolov5 
目前没有，动态输入涉及的更改项较多，当时在提供demo的时候，一方面是精力有限，一方面是为了以后维护相对简单些
最高只能32，分辨率320