parameters setting on busi dataset

[xiaohancl]

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]

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

[xiaohancl]

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

[jeya-maria-jose]

I just picked the first mask i.e _mask

[xiaohancl]

how about the parameters setting?running many times but i can't get the result as the same as it in your paper. Sorry to bother you!

---Original--- From: "Jeya Maria @.> Date: Sat, Apr 30, 2022 08:02 AM To: @.>; Cc: @.@.>; Subject: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busidataset (Issue #14)

I just picked the first mask i.e _mask

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

[jeya-maria-jose]

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]

channel setting i try the code you share(UNeXt-S 8 16 32 64 128) and then set the channel to [16 32 128 160 256]

---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: @.***>

[jeya-maria-jose]

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

[xiaohancl]

i will try again by changing the random seed.(i didn't change the random seed before)

---Original--- From: "Jeya Maria @.> Date: Sat, Apr 30, 2022 08:17 AM To: @.>; Cc: @.@.>; Subject: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busidataset (Issue #14)

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

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

[xiaohancl]

how about take the all masks into account?

---Original--- From: "Jeya Maria @.> Date: Sat, Apr 30, 2022 08:02 AM To: @.>; Cc: @.@.>; Subject: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busidataset (Issue #14)

I just picked the first mask i.e _mask

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

[liuyx599]

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]

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]

i tried another random seed and the result can arrive 66－68.You can change the random seed and run again.

---Original--- From: @.> Date: Thu, May 12, 2022 10:07 AM To: @.>; Cc: @.@.>; Subject: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busidataset (Issue #14)

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

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

[liuyx599]

i used the same setting as busi. On the GLAS, i splited the trainset in 8/2 then tested the whole testset, you can also try it

---Original--- From: @.> Date: Thu, May 12, 2022 10:12 AM To: @.>; Cc: @.@.>; Subject: Re: [jeya-maria-jose/UNeXt-pytorch] parameters setting on busi dataset (Issue #14)

我尝试了另一个随机种子，结果可以到达66－68。您可以更改随机种子并再次运行。 … @.> 日期：2022 年 5 月 12 日星期四上午 10:07 @.>; @.@.>; 主题：回复：busidataset 上的 [jeya-maria-jose/UNeXt-pytorch] 参数设置（问题#14）我在 busi 数据集上的表现也很差。是否要处理数据集？什么样的治疗？拿到数据集后，直接导入运行——直接回复这封邮件，在 GitHub 上查看，或者退订。您收到此消息是因为您编写了该线程。消息 @.***>

Can you show the code in detail? thank you.

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

[xiaohancl]

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]

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]

so that was it.I am sorry

---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: @.***>

[sdlymywr]

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

[liuyx599]

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]

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]

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: @.***>

[Chikeee]

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]

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: @.***>

[Y-Miou]

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]

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]

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]

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: @.***>

[YHYeooooong]

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]

@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]

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]

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: @.***>