search

Li-yachuan / NBED

Code of paper "A new baseline for edge detection: Make Encoder-Decoder great again"
31 stars 4 forks source link

pretrained weights #4

Open yx-yyds opened 1 month ago

yx-yyds commented 1 month ago

你好，可以提供一下在各个数据集上的预训练权重吗

Li-yachuan commented 1 month ago

不好意思，我之前的服务器暂时没有办法连接，你可以试着自己训练一下，很快的，应该是用不上10个小时就可以完成任意一个数据集上的训练

---Original--- From: @.> Date: Tue, Oct 22, 2024 14:09 PM To: @.>; Cc: @.***>; Subject: [Li-yachuan/NBED] pretrained weights (Issue #4)

你好，可以提供一下在各个数据集上的预训练权重吗

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

yx-yyds commented 1 month ago

好的，谢谢，我已经训练出来了

Li-yachuan commented 1 month ago

可以复现出我报告的精度吧？

---Original--- From: @.> Date: Thu, Oct 24, 2024 16:25 PM To: @.>; Cc: @.**@.>; Subject: Re: [Li-yachuan/NBED] pretrained weights (Issue #4)

好的，谢谢，我已经训练出来了

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

yx-yyds commented 1 month ago

请问在BSDS上是训练多少个epoch啊，我按你代码中的训练了10epoch，得到的单尺度测试结果ODS:0.832,OIS:0.839,不过我发现训练十个epoch损失函数并没有收敛，后面我又训练了20个epoch，发现损失函数依然还没收敛

Li-yachuan commented 1 month ago

就按照我给的那个训就行，训多了就过拟合了，你把每一个epoch的结果都测一下，最好的精度应该不是最后一个。

---Original--- From: @.> Date: Fri, Oct 25, 2024 08:06 AM To: @.>; Cc: @.**@.>; Subject: Re: [Li-yachuan/NBED] pretrained weights (Issue #4)

请问在BSDS上是训练多少个epoch啊，我按你代码中的训练了10epoch，得到的单尺度测试结果ODS:0.832,OIS:0.839,不过我发现训练十个epoch损失函数并没有收敛，后面我又训练了20个epoch，发现损失函数依然还没收敛

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

yx-yyds commented 1 month ago

啊，那在BIPED上也是这样吗

Li-yachuan commented 1 month ago

BIPED也有这种情况，相对会好一点

---Original--- From: @.> Date: Fri, Oct 25, 2024 08:28 AM To: @.>; Cc: @.**@.>; Subject: Re: [Li-yachuan/NBED] pretrained weights (Issue #4)

啊，那在BIPED上也是这样吗

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

yx-yyds commented 1 month ago

好的，

yx-yyds commented 1 month ago

你知道是什么原因导致这种情况发生吗，之前我也遇到过，一直没搞懂

Li-yachuan commented 1 month ago

模型的特征提取能力太强了，数据太少了，这是个正常现象。还发明了一个专门的名词来解决这个问题 叫early stop

---Original--- From: @.> Date: Fri, Oct 25, 2024 08:30 AM To: @.>; Cc: @.**@.>; Subject: Re: [Li-yachuan/NBED] pretrained weights (Issue #4)

你知道是什么原因导致这种情况发生吗，之前我也遇到过，一直没搞懂

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

yx-yyds commented 1 month ago

好的，谢谢