Toybrick

楼主: hisping

人工智能开发系列(3) YOLOV3开发与实现

 

rakidis

高级会员

积分
507
楼主
发表于 2020-4-14 16:21:07 | 显示全部楼层
Hello nin hao, sorry my chineese is not clear so I will be typing in english.

I receive the following error when I try to convert resnext50-yolov3 (608x608) model to rknn.

E ValueError: cannot reshape array of size 494399 into shape (1024,1024,1,1)
done

I realized error occurs during conversion to tensorflow, but don't know how to solve the issue.
Here is the cfg and model file for my case.

https://raw.githubusercontent.co ... riginal-optimal.cfg
https://drive.google.com/open?id ... WLNoXV8Mo7WKgwdYZCc

Thanks in advance, Regards
回复

使用道具 举报

rakidis

高级会员

积分
507
沙发
发表于 2020-4-20 16:43:39 | 显示全部楼层
I actually found out the reason why it wasn't able to convert that network. It turns out rknn toolkit can't convert depthwise convolution layers.
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则

产品中心 购买渠道 开源社区 Wiki教程 资料下载 关于Toybrick


快速回复 返回顶部 返回列表