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

输入矩形图片时检测框偏移 #30

Open
1 task done
wubalahao opened this issue Dec 26, 2021 · 9 comments
Open
1 task done

输入矩形图片时检测框偏移 #30

wubalahao opened this issue Dec 26, 2021 · 9 comments
Labels
question Further information is requested

Comments

@wubalahao
Copy link

Search before asking

Question

image

Additional

No response

@wubalahao wubalahao added the question Further information is requested label Dec 26, 2021
@acai66
Copy link
Owner

acai66 commented Dec 28, 2021

是指矩形输入训练吗

@wubalahao
Copy link
Author

wubalahao commented Dec 28, 2021 via email

@acai66
Copy link
Owner

acai66 commented Dec 28, 2021

直接用矩形原图,训练时网络输入尺寸用正方形,不影响最终的部署推理。

@wubalahao
Copy link
Author

好吧我刚才描述的不清楚。就是如果我用正方形的图片送入网络训练就没有问题,但是当我改成矩形的图片送入网络的时候标注框以及最终的检测框就会被拉扁,你之前说了代码改进后可以输入矩形了所以我没有很仔细的看您的代码,所以我也不知道你在矩形图片送入网络变成正方形的过程中有没有对标签数据也进行处理,导致了最后检测的时候检测框是扁的。

@acai66
Copy link
Owner

acai66 commented Dec 28, 2021

确保你的数据集标签是正确的就行,可以用原始xml标注格式的软件labelimg2、rolabelimg查看。
这个源码自始至终都可以用矩形训练,之前的偏移问题是数据增强时裁剪造成的,现在理论上不会有偏移问题了

@wubalahao
Copy link
Author

wubalahao commented Dec 28, 2021 via email

@wubalahao
Copy link
Author

d7b1cf94631f6cd57c6e4d385c14b06

484c4b6ddc4a88d2c82a764e22db40d

@acai66
Copy link
Owner

acai66 commented Jan 4, 2022

d7b1cf94631f6cd57c6e4d385c14b06 484c4b6ddc4a88d2c82a764e22db40d

还是建议检查下标签信息是否正确,看图里边框情况:其实中心点位置(以及角度)并未偏移,只是长度发生了变化,真正的偏移是中心点位置变了。

@tourhua
Copy link

tourhua commented Dec 2, 2022

想请问中心点发生了偏离是什么情况?预测框都往目标左上角偏了一点QAQ

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

No branches or pull requests

3 participants