RuntimeError: shape '[1, 20, 7, 35, 7, 256]' is invalid for input of size 8843520

[kingoo123]

Hi,I'm trying to test on my own test set, but the following issues arise and I would like to ask you how to fix them
The environment used is as follows：
python 3.8
cuda 10.1
pytorch 1.8.1
mmcv 1.4.8

[weihaosky]

Maybe related to the image size of your own data

[lcc157]

So，how to test own data?

[BinZhu-ece]

So，how to test own data?
I also want to know it.

[weihaosky]

It can be directly used on your own data. But you need to fix the bug.

[BinZhu-ece]

What I want to know is how to fix the bug, thank you!

…

------------------ 原始邮件 ------------------ 发件人: "Weihao ***@***.***>; 发送时间: 2022年8月9日(星期二) 上午10:42 收件人: ***@***.***>; 抄送: ***@***.***>; ***@***.***>; 主题: Re: [aliyun/NeWCRFs] RuntimeError: shape '[1, 20, 7, 35, 7, 256]' is invalid for input of size 8843520 (Issue #4) It can be directly used on your own data. But you need to fix the bug. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[754467737]

The reason for the error is that your image H,W is not divisible by 32 during the network downsampling process. you can refer to the yolo-v5--def letterbox to change the H,W to be divisible by 32.

[6leet]

Hi, I'm facing this problem currently, is there any general solution yet? Thanks!

[Ericchen328]

is there any solution to this problem