Zheng222 / IMDN

Lightweight Image Super-Resolution with Information Multi-distillation Network (ACM MM 2019, Winner Award of ICCVW AIM 2019 Constrained SR Track1&Track2)
418 stars 69 forks source link

关于验证与测试psnr不同的问题 #33

Open q623 opened 2 years ago

q623 commented 2 years ago

你好,为什么训练时候验证的psnr和test_imdn得出的psnr在同一个测试集上值是不同的呢

ziqiang00 commented 1 year ago

你好,请问你找到原因了吗?

q623 commented 1 year ago

没有 后面换 edsr 的框架来训练了

在 2022年8月31日,15:19,ziqiang00 @.***> 写道:

 你好,请问你找到原因了吗?

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.

ziqiang00 commented 1 year ago

我训练原网络模型也达不到他论文中的指标,你的可以吗

q623 commented 1 year ago

我最后也没去复现了😂

在 2022年9月4日,14:51,ziqiang00 @.***> 写道:

 我训练原网络模型也达不到他论文中的指标,你的可以吗

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.

ziqiang00 commented 1 year ago

hhhh,好吧

wwang0107 commented 1 year ago

xd,我看他的readme上面个训练指令都有加上预训练模型,作者在回答的时候也说多训练场几次,没准是不是把我们自己训练出来的那个模型作为预训练模型,多训练几次就能达到他的指标?

q623 commented 1 year ago

应该可以吧,兄弟可以试试

------------------ 原始邮件 ------------------ 发件人: "Zheng222/IMDN" @.>; 发送时间: 2022年9月13日(星期二) 上午9:38 @.>; @.**@.>; 主题: Re: [Zheng222/IMDN] 关于验证与测试psnr不同的问题 (Issue #33)

xd,我看他的readme上面个训练指令都有加上预训练模型,作者在回答的时候也说多训练场几次,没准是不是把我们自己训练出来的那个模型作为预训练模型,多训练几次就能达到他的指标?

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