2345678y3

> 您的图与代码不一致。图中有 5 个阶段，而代码中只有 4 个阶段。此外，下采样应先于 UX-Net 模块。 debug了下 图片是32倍下采样，但是他实际只有16倍下采样，原版3dunet也是只有16倍

My dataset is segmentation of brain metastases, and my labels may have 1 2 3 ... The maximum number of labels is 42, the size of the label voxels is...

> I have encountered your problem before, can you describe your situation in detail.↳ What is the reason for your situation?thank you

I have tried the nnUNetTrainerDiceCELoss_noSmooth but the problem hasn't been solved yet.

杨钦 ***@***.***  this is my training code and the result,thank you  ------------------ 原始邮件 ------------------ 发件人: "MIC-DKFZ/nnUNet" ***@***.***>; 发送时间: 2024年4月2日(星期二) 下午4:24 ***@***.***>; ***@***.******@***.***>; 主题: Re: [MIC-DKFZ/nnUNet] why the dice is 0 or nan (Issue #2034)...

> Hey, > > You did not send any additional information. Please review your reply again. Thank you. > > Best regards, Karol i have sent it to you again，could...

> Hey, > > You did not send any additional information. Please review your reply again. Thank you. > > Best regards, Karol [mynnuet.zip](https://github.com/MIC-DKFZ/nnUNet/files/14914347/mynnuet.zip)

Thank you for your answer，sir .My dataset is brain metastasis ,so,one patient  may have more than one metastasises that from the same tumor in his brain,thus,if the person has 42...

thank you for your help  sir 发自我的iPhone ------------------ Original ------------------ From: Karol Gotkowski ***@***.***> Date: Mon,May 6,2024 8:47 PM To: MIC-DKFZ/nnUNet ***@***.***> Cc: 2345678y3 ***@***.***>, Mention ***@***.***> Subject: Re: [MIC-DKFZ/nnUNet]...