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

parameters setting on busi dataset #14

Open
xiaohancl opened this issue Apr 28, 2022 · 46 comments
Open

parameters setting on busi dataset #14

xiaohancl opened this issue Apr 28, 2022 · 46 comments

Comments

@xiaohancl
Copy link

xiaohancl commented Apr 28, 2022

Sorry for accidentally turning this off!
My question is I set the parameters on busi dataset as fllows:
epoch:400
batch_size:8
optimizer:Adam
lr:1e-4
momentum:0.9
weight_decay:1e-4
scheduler:ConsineAnnealingLR
channels:[16, 32, 128, 160, 256]

I have resized the images by train_transform and val_transform you provided in the source code.And download the dataset from the link you provided, and then put the benign and malignant (total 647 images)in the folder inputs/busi/images, and the masks in inputs/busi/masks/0

but i got the result IOU:61. (it's 66.95 in the paper)

@xiaohancl
Copy link
Author

Maybe there is a problem with my dataset, because some pictures have more than one mask, how do you deal with it?

@xiaohancl
Copy link
Author

The effect is even worse when multiple masks are taken into account

@jeya-maria-jose
Copy link
Owner

I just picked the first mask i.e _mask

@xiaohancl
Copy link
Author

xiaohancl commented Apr 30, 2022 via email

@jeya-maria-jose
Copy link
Owner

I have used the same parameter settings you have specified. Can you please make sure you are using the model after this commit . The previous model file was actually UNeXt-S

@xiaohancl
Copy link
Author

xiaohancl commented Apr 30, 2022 via email

@jeya-maria-jose
Copy link
Owner

Maybe change the random seed and check average over different train-test splits?

@xiaohancl
Copy link
Author

xiaohancl commented Apr 30, 2022 via email

@xiaohancl
Copy link
Author

xiaohancl commented Apr 30, 2022 via email

@liuyx599
Copy link

liuyx599 commented May 3, 2022

I encountered the same situation, I used the original author's code with the same hyperparameters as the source code, trained 500 epochs on the BUSI dataset, and the highest IoU on the validation set could only reach about 61.3, and the dice score was about 75.2. Not as high as reported in the paper .... But I divided the training set 8/2 into training and validation sets on the GlaS dataset, and the highest IoU reached 79 on the validation set and 74.51 on the test set, which is higher than the 69.61 reported by your MedT model

@sdtegaotian
Copy link

I also performed poorly on the busi dataset. Do you want to process the data set? What kind of treatment? After I got the dataset, I directly imported it and ran it

@xiaohancl
Copy link
Author

xiaohancl commented May 12, 2022 via email

@liuyx599
Copy link

liuyx599 commented May 12, 2022 via email

@xiaohancl
Copy link
Author

This is not possible, because your test set of data in the training set appeared, the test effect must be good. The training set and the test set can't intersect

@liuyx599
Copy link

Sorry, maybe I misrepresented it. the GLAS dataset has a training set and a test set, I just divided the training set 8/2 into a training set and a validation set, and used the data from the test set during the testing phase, not the validation set which belongs to the 20% of the training set.

@xiaohancl
Copy link
Author

xiaohancl commented May 12, 2022 via email

@sdlymywr

This comment was marked as abuse.

@liuyx599
Copy link

I have a problem? By different random seeds, do you mean the parameters set in the initialization of the model, or others? For example, the following code

import numpy as np import torch import random import os seed_ Set value = 2020 # random number np. random. seed(seed_value) random. seed(seed_value) os. Environ ['pythonhashseed '] = str (seed_value) # to prohibit hash randomization and make the experiment repeatable. torch. manual_ Seed (seed_value) # sets a random seed for the CPU torch. cuda. manual_ Seed (seed_value) # sets a random seed for the current GPU (only one GPU) torch. cuda. manual_ seed_ All (seed_value) # set random seed for all GPUs (multiple GPUs) torch. backends. cudnn. deterministic = True

Can you share your code details

I used the author's latest arch.py submission, and with all other parameters the same, the validation set IoU reached 65.8, which is within the range reported in its paper

@xiaohancl
Copy link
Author

This is the source code provided by the author, just modifies the random_state parameter value during data partitioning.
train_img_ids, val_img_ids = train_test_split(img_ids, test_size=0.2, random_state=41)

@liuyx599
Copy link

liuyx599 commented May 12, 2022 via email

@Chikeee
Copy link

Chikeee commented Jun 7, 2022

就是这样。对不起

---Original--- From: @.> Date: Thu, May 12, 2022 10:26 AM To: @.>; Cc: @.@.>; Subject: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14) Sorry, maybe I misrepresented it. the GLAS dataset has a training set and a test set, I just divided the training set 8/2 into a training set and a validation set, and used the data from the test set during the testing phase, not the validation set which belongs to the 20% of the training set. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

I want to know the random state you set, my IOU cannot reach 66-68,thanks

@xiaohancl
Copy link
Author

xiaohancl commented Oct 11, 2022 via email

@Y-Miou
Copy link

Y-Miou commented Oct 30, 2022

我将发送我今天使用的代码。如果方便,您可以检查一下。请打扰您检查电子邮件。 再次感谢。
...
---源语言--- 寄件人:“杰亚·玛丽亚 @.> 日期: 2022年4月30日星期六上午08:09 收件人: @.>; 抄送: @.@.>; 主题:回复:busi 数据集上的 [jeya-maria-jose/UNeXt-pytorch] 参数设置(问题 #14) 我使用了您指定的相同参数设置。您能否确保在此提交后使用该模型.以前的模型文件实际上是UNeXt-S — 直接回复此电子邮件,在 GitHub 上查看或取消订阅。 您收到此消息是因为您创作了线程。消息 ID: @.**>

亲。能发我一下您修改的达到66-68的效果的代码吗?非常感谢您的帮助! @小汉科

@Y-Miou
Copy link

Y-Miou commented Oct 30, 2022

我将发送我今天使用的代码。如果方便,您可以检查一下。请打扰您检查电子邮件。 再次感谢。
...
---源语言--- 寄件人:“杰亚·玛丽亚 @.> 日期: 2022年4月30日星期六上午08:09 收件人: @.>; 抄送: @.@.>; 主题:回复:busi 数据集上的 [jeya-maria-jose/UNeXt-pytorch] 参数设置(问题 #14) 我使用了您指定的相同参数设置。您能否确保在此提交后使用该模型.以前的模型文件实际上是UNeXt-S — 直接回复此电子邮件,在 GitHub 上查看或取消订阅。 您收到此消息是因为您创作了线程。消息 ID: @.**>
Hello, the two data sets I trained have not reached the effect in the paper, could you please send me how you modified them? My email is [email protected]. Thank you very much for your help!

@YHYeooooong
Copy link

YHYeooooong commented Dec 20, 2022

I will send the code i used within today.If convenience, you can check it. Please bother you to check the email. Thanks again.

---Original--- From: "Jeya Maria @.> Date: Sat, Apr 30, 2022 08:09 AM To: @.>; Cc: @.@.>; Subject: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14) I have used the same parameter settings you have specified. Can you please make sure you are using the model after this commit . The previous model file was actually UNeXt-S — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

Hello, I have the same problem for reproduced the 66-68 IOU score based on the BUSI dataset. Could you send me the random state value you used? My email is [email protected] . Thanks!

@Y-Miou
Copy link

Y-Miou commented Dec 20, 2022 via email

@YHYeooooong
Copy link

非常感谢您的帮助,我很需要您的代码,希望您能分享一下。  

------------------ 原始邮件 ------------------ 发件人: "jeya-maria-jose/UNeXt-pytorch" @.>; 发送时间: 2022年12月20日(星期二) 上午10:47 @.>; @.@.>; 主题: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14) I will send the code i used within today.If convenience, you can check it. Please bother you to check the email. Thanks again. … ---Original--- From: "Jeya Maria @.> Date: Sat, Apr 30, 2022 08:09 AM To: @.>; Cc: @.@.>; Subject: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14) I have used the same parameter settings you have specified. Can you please make sure you are using the model after this commit . The previous model file was actually UNeXt-S — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.> Hello, I have the same problem for reproduced the 66-68 IOU score based on the BUSI dataset. Could you send me the random state value you used? thanks!! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

Hi, I did not get the code for the random state values, but I tried some random state values. In my case, the random state value (46) leads to the highest val IoU score of 0.668 .

@aswa123
Copy link

aswa123 commented Feb 23, 2023

@jeya-maria-jose @YHYeooooong @sdlymywr @sdtegaotian @xiaohancl @liuyx599

ValueError: With n_samples=0, test_size=0.2 and train_size=None, the resulting train set will be empty. Adjust any of the aforementioned parameters.

I got this error while using busi dataset.
how can I adjust setting?
please help!

@JiuZhouu
Copy link

JiuZhouu commented Apr 9, 2023

非常感谢您的帮助,我很需要您的代码,希望您能分享一下。  

------------------ 原始邮件 ------------------ 发件人: "jeya-maria-jose/UNeXt-pytorch" @.>; 发送时间: 2022年12月20日(星期二) 上午10:47 _@**._>; _@.@.>; 主题: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14) I will send the code i used within today.If convenience, you can check it. Please bother you to check the email. Thanks again. … ---Original--- From: "Jeya Maria @.> Date: Sat, Apr 30, 2022 08:09 AM To: @.>; Cc: @.@.>; Subject: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14) I have used the same parameter settings you have specified. Can you please make sure you are using the model after this commit . The previous model file was actually UNeXt-S — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.> Hello, I have the same problem for reproduced the 66-68 IOU score based on the BUSI dataset. Could you send me the random state value you used? thanks!! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.**_>

Hi, I did not get the code for the random state values, but I tried some random state values. In my case, the random state value (46) leads to the highest val IoU score of 0.668 .

also used 46 and got better result, Thank you !

@sdtegaotian
Copy link

sdtegaotian commented Apr 9, 2023 via email

@sdtegaotian
Copy link

sdtegaotian commented Apr 9, 2023 via email

@Cloud-Liu
Copy link

Hello, did you run the metrics in the paper, I set the same parameters and couldn't achieve the results in the paper

@112299aa
Copy link

I will send the code i used within today.If convenience, you can check it. Please bother you to check the email. Thanks again.

---Original--- From: "Jeya Maria @.> Date: Sat, Apr 30, 2022 08:09 AM To: _@**._>; Cc: _@.@._>; Subject: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14) I have used the same parameter settings you have specified. Can you please make sure you are using the model after this commit . The previous model file was actually UNeXt-S — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: _@_.*>

Hello, I have the same problem for reproduced the 66-68 IOU score based on the BUSI dataset. Could you send me the random state value you used? My email is [email protected] . Thanks!

Can you send me your code?Because my result is poor.Even though I set it 46.But i need set the val.py 46.Then the result is poor

@112299aa
Copy link

非常感谢您的帮助,我很需要您的代码,希望您能分享一下。  

------------------ 原始邮件 ------------------ 发件人: "jeya-maria-jose/UNeXt-pytorch" @.>; 发送时间: 2022年12月20日(星期二) 上午10:47 _@**._>; _@.@.>; 主题: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14) I will send the code i used within today.If convenience, you can check it. Please bother you to check the email. Thanks again. … ---Original--- From: "Jeya Maria @.> Date: Sat, Apr 30, 2022 08:09 AM To: @.>; Cc: @.@.>; Subject: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14) I have used the same parameter settings you have specified. Can you please make sure you are using the model after this commit . The previous model file was actually UNeXt-S — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.> Hello, I have the same problem for reproduced the 66-68 IOU score based on the BUSI dataset. Could you send me the random state value you used? thanks!! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.**_>

Hi, I did not get the code for the random state values, but I tried some random state values. In my case, the random state value (46) leads to the highest val IoU score of 0.668 .

also used 46 and got better result, Thank you !

Is your val.py also set 46.I do it,but I got a poor result.Can you send me your code?

@112299aa
Copy link

Chinese

---- 回复的原邮件 ---- | 发件人 | @.> | | 日期 | 2023年04月09日 18:30 | | 收件人 | @.> | | 抄送至 | @.>@.> | | 主题 | Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14) | 非常感谢您的帮助,我很需要您的代码,希望您能分享一下。 … ------------------ 原始邮件 ------------------ 发件人: "jeya-maria-jose/UNeXt-pytorch" @.>; 发送时间: 2022年12月20日(星期二) 上午10:47 @**.>; @.@.>; 主题: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14) I will send the code i used within today.If convenience, you can check it. Please bother you to check the email. Thanks again. … ---Original--- From: "Jeya Maria @.> Date: Sat, Apr 30, 2022 08:09 AM To: @.>; Cc: @.@.>; Subject: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14) I have used the same parameter settings you have specified. Can you please make sure you are using the model after this commit . The previous model file was actually UNeXt-S — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.> Hello, I have the same problem for reproduced the 66-68 IOU score based on the BUSI dataset. Could you send me the random state value you used? thanks!! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @._> Hi, I did not get the code for the random state values, but I tried some random state values. In my case, the random state value (46) leads to the highest val IoU score of 0.668 . also used 46 and got better result, Thank you ! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.**>

I am chinese too.你复现出和作者差不多的结果了吗?可以分享一下你的代码吗?因为我把训练和验证的 random state都设置46了,结果还是比较差

@112299aa
Copy link

Would you please share you code with me.my result is poor.I am sorry to bother you.

@112299aa
Copy link

112299aa commented Jun 1, 2023

@jeya-maria-jose @YHYeooooong @sdlymywr @sdtegaotian @xiaohancl @liuyx599

ValueError: With n_samples=0, test_size=0.2 and train_size=None, the resulting train set will be empty. Adjust any of the aforementioned parameters.

I got this error while using busi dataset. how can I adjust setting? please help!

you dataset name maybe worry

@112299aa
Copy link

112299aa commented Jun 1, 2023

I just used the same code, same random state and reached 65.8 on busi

---Original--- From: @.> Date: Thu, May 12, 2022 11:07 AM To: @.>; Cc: @.@.>; Subject: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busidataset (Issue #14) This is the source code provided by the author, just modifies the random_state parameter value during data partitioning. train_img_ids, val_img_ids = train_test_split(img_ids, test_size=0.2, random_state=41) — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

Could you please share your code? Because mine is less than 65.I am very worried

@sdtegaotian
Copy link

sdtegaotian commented Jun 18, 2023 via email

@Cloud-Liu
Copy link

yes ,I also can't achieve this result.But I make some change.With this change I can get better results.So I make some advice?with you that using this code to add yourself code.

---- Replied Message ---- | From | @.> | | Date | 05/30/2023 23:07 | | To | @.> | | Cc | @.>@.> | | Subject | Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14) | Hello, did you run the metrics in the paper, I set the same parameters and couldn't achieve the results in the paper — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

你好,可以发我一份代码吗?[email protected]

@Y-Miou
Copy link

Y-Miou commented Jun 18, 2023 via email

@Y-Miou
Copy link

Y-Miou commented Jun 18, 2023 via email

@112299aa
Copy link

112299aa commented Jun 18, 2023 via email

@wangx0530
Copy link

yes ,I also can't achieve this result.But I make some change.With this change I can get better results.So I make some advice?with you that using this code to add yourself code.

---- Replied Message ---- | From | @.> | | Date | 05/30/2023 23:07 | | To | _@**._> | | Cc | _@.>@._> | | Subject | Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14) | Hello, did you run the metrics in the paper, I set the same parameters and couldn't achieve the results in the paper — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: _@_.*>

你好,可以发我一份代码吗?[email protected]

你好,我想问一下这个BUSI数据集 是将647张图片混到一个文件夹中训练吗

@wangx0530
Copy link

wangx0530 commented Jan 17, 2024

你好,我想问一下这个BUSI数据集 是将647张图片混到一个文件夹中训练吗

@112299aa
Copy link

112299aa commented Jan 17, 2024 via email

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