Pytorch-UNet icon indicating copy to clipboard operation
Pytorch-UNet copied to clipboard

训练结果不好,怎么调参数

Open 770338417lyy opened this issue 4 years ago • 10 comments

info:validation dice coeff:0.14 左右波动,,,请问大佬些该怎么调参数

770338417lyy avatar Nov 09 '21 06:11 770338417lyy

用了200张图片,跑了200轮,训练集当测试集,效果如下。不知道那个环节出了问题。 image

GanHey avatar Nov 10 '21 01:11 GanHey

不知道那个环节出了问题

GanHey avatar Nov 10 '21 01:11 GanHey

image The loss is strange,and the learning rate has dropped to 0

GanHey avatar Nov 10 '21 03:11 GanHey

metoo,我觉得可能是数据集的问题

asako98 avatar Nov 25 '21 08:11 asako98

image The loss is strange,and the learning rate has dropped to 0

hello,had you solved this question?

Jazmineyjy avatar May 25 '22 13:05 Jazmineyjy

metoo,我觉得可能是数据集的问题

你好,请问你解决这个问题了吗,我也遇到了同样的问题

Jazmineyjy avatar May 25 '22 13:05 Jazmineyjy

image The loss is strange,and the learning rate has dropped to 0 你好请问你有啥解决办法了嘛,我的也是效果不是很好,损失降不下去

mmmmyolo avatar Nov 22 '22 04:11 mmmmyolo

我换了其他的模型训练的,不好意思,没有帮助到你

---原始邮件--- 发件人: @.> 发送时间: 2022年11月22日(周二) 中午12:03 收件人: @.>; 抄送: @.@.>; 主题: Re: [milesial/Pytorch-UNet] 训练结果不好,怎么调参数 (Issue #314)

The loss is strange,and the learning rate has dropped to 0 你好请问你有啥解决办法了嘛,我的也是效果不是很好,损失降不下去

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

770338417lyy avatar Nov 22 '22 14:11 770338417lyy

可能是数据集差异太大,预处理不够好,模型学不到记不住,先用几张图试试能不能过拟合

tensorctn avatar Jan 13 '24 02:01 tensorctn