search

Xinyu-Xiang / DIVFusion

DIVFusion: Darkness-free infrared and visible image fusion
106 stars 10 forks source link

代码运行损失值出现nan #6

Open hyl510 opened 1 year ago

hyl510 commented 1 year ago

您好，我按照readme的步骤下载了vgg16和train文件（将240对照片分成200对训练集和40对测试集），但是损失值一直是nan，请问是什么原因呢？

Xinyu-Xiang commented 1 year ago

非常抱歉，这个我在训练或者测试的时候没有出现过这个问题诶，按道理来说应该可以直接训练的，会不会是一些路径问题或者图像数据未载入之类的

seaxia commented 1 year ago

您好，我按照readme的步骤下载了vgg16和train文件（将240对照片分成200对训练集和40对测试集），但是损失值一直是nan，请问是什么原因呢？

Hello, is your problem solved? I also encountered the same problem.

hyl510 commented 1 year ago

     No, it didn't. I ran other codes to complete my homework. (Sometimes we have to learn to give up.)       It turned Nan as soon as the training started, but researchers said that they haven't encountered the problem like this. I guess it may have something wrong with the datasets they provided? You can try to obtain the original datasets to see whether it works. Just a piece of advice...       Wish you to complete your work successfully！

  发自我的企业微信

----------回复的邮件信息---------- @.***> 于2023年6月6日(周二) 下午3:05写道:

您好，我按照readme的步骤下载了vgg16和train文件（将240对照片分成200对训练集和40对测试集），但是损失值一直是nan，请问是什么原因呢？

Hello, is your problem solved? I also encountered the same problem.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

Xinyu-Xiang commented 1 year ago

会不会是因为tf版本中部分函数不配，所以计算中出现的问题呢，因为我这边确实没有遇到过nan的情况

Haolin0817 commented 3 months ago

您好，我按照readme的步骤下载了vgg16和train文件（将240对照片分成200对训练集和40对测试集），但是损失值一直是nan，请问是什么原因呢？

您好，您的问题解决了吗？我也遇到了同样的问题。