Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

About loss, image_loss when train model? #16

Open
vvthai10 opened this issue May 25, 2024 · 42 comments
Open

About loss, image_loss when train model? #16

vvthai10 opened this issue May 25, 2024 · 42 comments

Comments

@vvthai10
Copy link

I train with dataset visa,
I train 11 epoch, but loss and image_loss is 3.7960, 0.5325. I feel something wrong. I setup by your setting. Can you share me about loss and image_loss when u train?

@wyy123de
Copy link

image
我的损失比图像损失还小

@vvthai10
Copy link
Author

vvthai10 commented May 28, 2024 via email

@vvthai10
Copy link
Author

and how can you show the table after train 15 epochs, I not find way to show this

@wyy123de
Copy link

image

@wyy123de
Copy link

这个你要自己保存到results中,我用的Head-CT数据集
还可以可视化呢

@wyy123de
Copy link

image

@vvthai10
Copy link
Author

You train by Brain dataset, not mvtec

@vvthai10
Copy link
Author

Can u share me code you visualize graphwhen train model

@wyy123de
Copy link

对啊,我是医学影响异常检测

@wyy123de
Copy link

sure

@vvthai10
Copy link
Author

I'm also studying the medical part like you, learnign_rate u still 0.001 ?

@wyy123de
Copy link

yes

@vvthai10
Copy link
Author

can u share me link of dataset u use,

@vvthai10
Copy link
Author

I try train with mvtec, after 15 epoch, loss, and image_loss is 3.4, 0.3 but when test result still okay

@wyy123de
Copy link

嗯嗯HeadCT

@wyy123de
Copy link

我感觉我训练有BUG

@wyy123de
Copy link

原文效果没那么好,哈哈哈哈

@wyy123de
Copy link

HeadCT

@wyy123de
Copy link

parser.add_argument("--train_data_path", type=str, default="/root/Downloads/Untitled Folder/AnomalyCLIP-main", help="train dataset path")
parser.add_argument("--save_path", type=str, default='/root/Downloads/Untitled Folder/AnomalyCLIP-main/results', help='path to save results')
这个是保存代码,其实作者给的代码很详细,你要保存到位置就可以看到

@vvthai10
Copy link
Author

And now i want to add adapter to image encoder, u want try with me

@vvthai10
Copy link
Author

"The original text is not that good" I don't understand what u say.

@wyy123de
Copy link

image
训练出现了BUG

@wyy123de
Copy link

导致效果好

@wyy123de
Copy link

image

@vvthai10
Copy link
Author

你的意思是你的结果比文章高很多吧?

@wyy123de
Copy link

@vvthai10
Copy link
Author

看起来他们在文章中使用 mvtec 进行训练并在医疗设备上进行评估,但如果你使用医疗包进行训练,我认为结果会有所不同。

@wyy123de
Copy link

image
跑了100个epoch

@vvthai10
Copy link
Author

为什么训练了 100 个 epoch,结果却低于 15 个 epoch?

@vvthai10
Copy link
Author

在数据集文件中,我没有看到的generate_class_info函数返回大脑数据集的obj_list

@wyy123de
Copy link

100个epoch,训练的多,我学习率调到0.000001,15个epoch学习率0.001

@vvthai10
Copy link
Author

我发现在这种类型的练习中,他们仅训练 15 或 50 个 epoch,lr 为 0.001 或 0.0001。 因为训练数据很小,训练多了就怕过拟合

@wyy123de
Copy link

yes

@vvthai10
Copy link
Author

但是你的损失确实很低,我不明白为什么,因为当我用mvtec文件训练时,它是3.x,哈哈哈

@vvthai10
Copy link
Author

您想与您的团队就这个主题进行合作吗? 我正在尝试更新图像编码器部分以将其合并到本文中。

@wyy123de
Copy link

可能你才是对的,不过损失不下降,可能是欠拟合,你可以尝试调大epoch,或者学习率

但是你的损失确实很低,我不明白为什么,因为当我用mvtec文件训练时,它是3.x,哈哈哈

@wyy123de
Copy link

wyy123de commented May 28, 2024 via email

@vvthai10
Copy link
Author

vvthai10 commented May 28, 2024 via email

@vvthai10
Copy link
Author

这里没有显示我的电子邮件地址。我的 Gmail 是 vuvanthai1410。

@wyy123de
Copy link

ok

@vvthai10
Copy link
Author

请给我发电子邮件,我会详细告诉你我在做什么:)

@vvthai10
Copy link
Author

yeah, i ssee

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants