search

huawei-noah / Efficient-AI-Backbones

Efficient AI Backbones including GhostNet, TNT and MLP, developed by Huawei Noah's Ark Lab.
4.07k stars 708 forks source link

question of test #145

Closed ModeSky closed 2 years ago

ModeSky commented 2 years ago

What causes the following errors in the test report? 1664332868085

iamhankai commented 2 years ago

I re-run the code of pvig_m_224_gelu, and succeed. Did you modify the code of pyramid_vig.py?

ModeSky commented 2 years ago

您好，我没有更改原模型代码部分

------------------ 原始邮件 ------------------ 发件人: "Kai @.>; 发送时间: 2022年9月28日(星期三) 下午2:37 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [huawei-noah/Efficient-AI-Backbones] question of test (Issue #145)

I re-run the code of pvig_m_224_gelu, and succeed. Did you modify the code of pyramid_vig.py?

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

ModeSky commented 2 years ago

作者您好，我想咨询一下，对于ImageNet数据集体量太大，本人将其缩减到了16万张数据，但使用ViG跑该数据集却显示train的loss一直处于6.5以上，然后test部分top1和5的准确率都很低，请问您能解答一下疑惑吗，谢谢

------------------ 原始邮件 ------------------ 发件人: "Kai @.>; 发送时间: 2022年9月28日(星期三) 下午2:37 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [huawei-noah/Efficient-AI-Backbones] question of test (Issue #145)

I re-run the code of pvig_m_224_gelu, and succeed. Did you modify the code of pyramid_vig.py?

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

iamhankai commented 2 years ago

请问怎么缩减的数据集？

train loss一直6.5以上，看起来是没收敛，需要训到3左右才好。

ModeSky commented 2 years ago

就是选择了test文件夹前10000张图片，train文件夹各子文件夹留前100张图片，val文件夹留取前50000张图片，然后剩下的ILSVRC2012_devkit_t12文件夹未更改

------------------ 原始邮件 ------------------ 发件人: "Kai @.>; 发送时间: 2022年9月28日(星期三) 下午3:16 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [huawei-noah/Efficient-AI-Backbones] question of test (Issue #145)

请问怎么缩减的数据集？

train loss一直6.5以上，看起来是没收敛，需要训到3左右才好。

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

iamhankai commented 2 years ago

那你还不如保留100类，每类100张可能有点少

ModeSky commented 2 years ago

保留100类的话，test文件夹里包含1000类数据，请问test文件夹如何处理呢

------------------ 原始邮件 ------------------ 发件人: "Kai @.>; 发送时间: 2022年9月28日(星期三) 下午3:36 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [huawei-noah/Efficient-AI-Backbones] question of test (Issue #145)

那你还不如保留100类，每类100张可能有点少

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

iamhankai commented 2 years ago

使用对应的100类做val和test

ModeSky commented 2 years ago

您好，请问完成这些操作后形成的数据集还需要生成对应的标签文件吗？例如class.txt和label.txt这种

------------------ 原始邮件 ------------------ 发件人: "huawei-noah/Efficient-AI-Backbones" @.>; 发送时间: 2022年9月28日(星期三) 下午3:49 @.>; @.**@.>; 主题: Re: [huawei-noah/Efficient-AI-Backbones] question of test (Issue #145)

使用对应的100类做val和test

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

ModeSky commented 2 years ago

作者您好，根据您的意思，我将val的100个文件夹及类留下后，要自己制作test集还是我不需留下test？

Liar. @.***

 

------------------ 原始邮件 ------------------ 发件人: "huawei-noah/Efficient-AI-Backbones" @.>; 发送时间: 2022年9月28日(星期三) 下午3:49 @.>; @.**@.>; 主题: Re: [huawei-noah/Efficient-AI-Backbones] question of test (Issue #145)

使用对应的100类做val和test

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

iamhankai commented 2 years ago

用不到test

ModeSky commented 2 years ago

好的，谢谢  

Liar. @.***

 

------------------ 原始邮件 ------------------ 发件人: "Kai @.>; 发送时间: 2022年9月29日(星期四) 上午10:57 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [huawei-noah/Efficient-AI-Backbones] question of test (Issue #145)

用不到test

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

ModeSky commented 2 years ago

作者您好，根据您的最近几天的解答，我是不是可以理解ViG这个网络架构是个无监督学习的呢？

Liar. @.***

 

？ ------------------ 原始邮件 ------------------ 发件人: "huawei-noah/Efficient-AI-Backbones" @.>; 发送时间: 2022年9月29日(星期四) 上午10:57 @.>; @.**@.>; 主题: Re: [huawei-noah/Efficient-AI-Backbones] question of test (Issue #145)

用不到test

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

ModeSky commented 2 years ago

作者您好，想问一下，本人打印train.py、myloader.py以及向上溯源的loader.py中打印target变量时显示如下，是因为未更改参数变量吗？

Liar. @.***

 

------------------ 原始邮件 ------------------ 发件人: "huawei-noah/Efficient-AI-Backbones" @.>; 发送时间: 2022年9月29日(星期四) 上午10:57 @.>; @.**@.>; 主题: Re: [huawei-noah/Efficient-AI-Backbones] question of test (Issue #145)

用不到test

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

iamhankai commented 2 years ago

没看到图片

ModeSky commented 2 years ago

Liar. @.***

 

------------------ 原始邮件 ------------------ 发件人: "huawei-noah/Efficient-AI-Backbones" @.>; 发送时间: 2022年9月30日(星期五) 上午10:52 @.>; @.**@.>; 主题: Re: [huawei-noah/Efficient-AI-Backbones] question of test (Issue #145)

没看到图片

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

ModeSky commented 2 years ago

作者您好，我刚刚训练一下代码，数据集是100类，但是目前来看好像不太对

Liar. @.***

 

------------------ 原始邮件 ------------------ 发件人: "huawei-noah/Efficient-AI-Backbones" @.>; 发送时间: 2022年9月30日(星期五) 上午10:52 @.>; @.**@.>; 主题: Re: [huawei-noah/Efficient-AI-Backbones] question of test (Issue #145)

没看到图片

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

ModeSky commented 2 years ago

作者您好，请问这种情况是数据集未处理好，导致程序未读取到标签还是myloader中的参数和我的环境有所差异，打印train中的target显示乱码，导致test时acc的值大于1，误差很大

Liar. @.***

 

------------------ 原始邮件 ------------------ 发件人: "huawei-noah/Efficient-AI-Backbones" @.>; 发送时间: 2022年9月30日(星期五) 上午10:52 @.>; @.**@.>; 主题: Re: [huawei-noah/Efficient-AI-Backbones] question of test (Issue #145)

没看到图片

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

ModeSky commented 2 years ago

好的，谢谢您

------------------ 原始邮件 ------------------ 发件人: "Kai @.>; 发送时间: 2022年9月28日(星期三) 下午3:49 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [huawei-noah/Efficient-AI-Backbones] question of test (Issue #145)

使用对应的100类做val和test

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

goodmorningsi commented 1 year ago

What causes the following errors in the test report? 1664332868085 您好，我也遇到了这个问题，请问这个问题要怎么解决呢？

goodmorningsi commented 1 year ago

请问楼主可以留个邮箱吗，有些问题想要向你请教