xuebinqin
U2Net在多类别语义分割中的应用问题
看了大佬的代码，最后的output和Loss均基于Sigmoid的二分类问题，如果要改成多类别的语义分割问题，网络的最后output和Loss是不是要重新设计，比如，我要分割图像中的person和dog，意味着网络的output的channel最后需要是2，sigmoid需要替换为softmax,相应的Loss也要替换为多分类的Loss是这样吗？
是的，output channel 修改成多通道，sigmoid修改成softmax, loss改成crossentropy.
On Nov 12, 2020, at 1:24 AM, 徐静 [email protected] wrote:
看了大佬的代码，最后的output和Loss均基于Sigmoid的二分类问题，如果要改成多类别的语义分割问题，网络的最后output和Loss是不是要重新设计，比如，我要分割图像中的person和dog，意味着网络的output的channel最后需要是2，sigmoid需要替换为softmax,相应的Loss也要替换为多分类的Loss是这样吗？
我有问题想请教，在修改为多分类是label中的图片应该是以什么形式存在的？灰度图像吗？
output channel 修改成多通道，sigmoid修改成softmax, loss改成crossentropy.这些部分我都修改了，使用多分类数据集，label为灰度图像，但是没有成功运行。
The dataloader has also to be modified for loading the multi-class masks. Different intteger values e.g. [0,n-1], where n denotes the number of classes, are usually used to represent different classes. You can also convert then to multi-channel one-hot format.
On Wed, Dec 23, 2020 at 5:29 AM taocx [email protected] wrote:
我有问题想请教，在修改为多分类是label中的图片应该是以什么形式存在的？灰度图像吗？
output channel 修改成多通道，sigmoid修改成softmax, loss改成crossentropy.这些部分我都修改了，使用多分类数据集，label为灰度图像，但是没有成功运行。
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/NathanUA/U-2-Net/issues/92#issuecomment-750260275, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGOROIED4S5FSIVGRYGGTSWHPBPANCNFSM4TS7ADRA .
-- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/
你好，你训练多类别分割成功了吗？可以分享下你修改的经验吗？除了output channel 修改成多通道，sigmoid修改成softmax, loss改成crossentropy，还有什么需要修改和注意的吗？
Hi, thanks for your interest. We haven't tried semantic tasks yet. But someone told us they tried and it works fine. I think to handle semantic segmentation, you probably have to increase the middle channel filter numbers as well as trying to remove some of the dense supervisions to obtain larger fitting capacities. Of course, these are just some possible ways, of which the performance may not be guaranteed.
On Wed, Jan 19, 2022 at 2:18 PM 曹小草 @.***> wrote:
你好，你训练多类别分割成功了吗？可以分享下你修改的经验吗？除了output channel 修改成多通道，sigmoid修改成softmax, loss改成crossentropy，还有什么需要修改和注意的吗？
— Reply to this email directly, view it on GitHub https://github.com/xuebinqin/U-2-Net/issues/92#issuecomment-1016287226, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORMWK5RNROEJZ2NTKWTUW2FWVANCNFSM4TS7ADRA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
You are receiving this because you commented.Message ID: @.***>
-- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage: https://xuebinqin.github.io/