关于数据集以及测试指标问题

[2805413893]

您好，我按照您论文中描述的进行数据集的预处理，进行训练，但是一直达不到您论文中提到的指标，差距比较大，请问您愿意分享您当时使用的数据集以及您的参数设置吗，感谢您的帮助！

[taozh2017]

你好，我先整理下数据，然后放到网盘里。另外，你说的差别很大 是有多大？可能与batchsize有关系，我当时有4块32G的GPU运行实验。

[2805413893]

指标差不多用您的源代码差不多是20.57，我改进了一些之后是21.68左右，一直超不过22，所以就很迷惑，我也怀疑跟batchsize有关，您的实验环境我目前还是达不到，希望可以用一下您的数据集试试，因为我处理的数据集存在那种全黑或者全白图像的情况不知道是否有影响，非常非常感谢您的帮助 发自我的iPhone

…

------------------ 原始邮件 ------------------ 发件人: Tao Zhou ***@***.***> 发送时间: 2021年10月4日 12:16 收件人: taozh2017/HiNet ***@***.***> 抄送: candy ***@***.***>, Author ***@***.***> 主题: 回复：[taozh2017/HiNet] 关于数据集以及测试指标问题 (#14) 你好，我先整理下数据，然后放到网盘里。另外，你说的差别很大 是有多大？可能与batchsize有关系，我当时有4块32G的GPU运行实验。 — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

[2805413893]

您好，我发现可能是由于我自己制作数据集的问题，希望可以使用您当时实验的数据集训练一下，感谢您的帮助！

[taozh2017]

我尽快哈，最近事情有点多。 candy ***@***.***> 于2021年10月6日周三 上午9:33写道：

…

您好，我发现可能是由于我自己制作数据集的问题，希望可以使用您当时实验的数据集训练一下，感谢您的帮助！ — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#14 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHT5QRJ2NUEST4SVNQXGIKDUFOROVANCNFSM5FIJTLHQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>.

-- Best, Tao.

[2805413893]

已经给您发送了邮件，请您查看一下，感谢

[Yshouang]

Can you share your data set? I have never run out the index of your paper, which is a little different. Besides, there are 3KB of data you put in the zip. Do you want to follow the training of these 3KB? Also, is it OK not to cut into 4 pieces of 128*128? Why did you write only one synthesized code in the synthesis section? Is there no synthesized code that can be put out for reference? Thank you

[2805413893]

I think the test indicators are different from those in the paper because the test methods are different. The author did not give the complete test code in the source code, and there is no code to generate the resulting image.

[Yshouang]

我认为测试指标与论文中的不同，因为测试方法不同。作者在源码中没有给出完整的测试代码，也没有生成结果图的代码。

Have you finally achieved the result? Why do I use the author's def prediction_syn_results function in utils.py Even though PSNR is around 21, the saved results have obvious stitching marks. Have you written the final result chart code? What's your email address, please? I want to learn from you. I am a new student. Thank you very much for your reply.

[2805413893]

There is a big gap between my test results and the paper. About 23 psnr, I didn’t write the code to generate the result image very well. I also hope to get the author’s source code. My email address is introduced on my github. In, we learn from each other, mainly hope that the original author can further answer

[taozh2017]

There is a big gap between my test results and the paper. About 23 psnr, I didn’t write the code to generate the result image very well. I also hope to get the author’s source code. My email address is introduced on my github. In, we learn from each other, mainly hope that the original author can further answer

I remembered I have sent the data to you, right?

[2805413893]

There is a big gap between my test results and the paper. About 23 psnr, I didn’t write the code to generate the result image very well. I also hope to get the author’s source code. My email address is introduced on my github. In, we learn from each other, mainly hope that the original author can further answer

I remembered I have sent the data to you, right?

Yes, I have processed your data set. Except for the psnr index, MSE and ssim are not much different from those in the paper, but it may be a problem with the test method. I test the index of each slice of the case and do it evenly. , And then average each case in the test set. At present, the average value is about 23. I think it is related to the batchsize. The batchsize of 16 is better than 4. I see that the use of 128 in your paper seems to be 128. At present, it is mainly Hope to get your guidance on the way of testing indicators, thank you for your continued help

[taozh2017]

There is a big gap between my test results and the paper. About 23 psnr, I didn’t write the code to generate the result image very well. I also hope to get the author’s source code. My email address is introduced on my github. In, we learn from each other, mainly hope that the original author can further answer

I remembered I have sent the data to you, right?

Yes, I have processed your data set. Except for the psnr index, MSE and ssim are not much different from those in the paper, but it may be a problem with the test method. I test the index of each slice of the case and do it evenly. , And then average each case in the test set. At present, the average value is about 23. I think it is related to the batchsize. The batchsize of 16 is better than 4. I see that the use of 128 in your paper seems to be 128. At present, it is mainly Hope to get your guidance on the way of testing indicators, thank you for your continued help

PSNR is also related to the size of the test image, e.g., PSNR of 160180 image is lower than PSNR of 256256 image. Besides, you can test your own model to see how about the PSNR?

[Yshouang]

I'm very sorry for not replying to your message in time. I've been busy with my class. I don't seem to have received the data set you sent me. I hope you can send me a new one. Thank you very much. Because I am also in the preliminary study stage, which is the first semester of my postgraduate career, I hope to keep communicating with you, thank you very much. It seems that the author didn't respond to these questions. At 2021-12-01 11:21:06, "candy" ***@***.***> wrote: There is a big gap between my test results and the paper. About 23 psnr, I didn’t write the code to generate the result image very well. I also hope to get the author’s source code. My email address is introduced on my github. In, we learn from each other, mainly hope that the original author can further answer I remembered I have sent the data to you, right? Yes, I have processed your data set. Except for the psnr index, MSE and ssim are not much different from those in the paper, but it may be a problem with the test method. I test the index of each slice of the case and do it evenly. , And then average each case in the test set. At present, the average value is about 23. I think it is related to the batchsize. The batchsize of 16 is better than 4. I see that the use of 128 in your paper seems to be 128. At present, it is mainly Hope to get your guidance on the way of testing indicators, thank you for your continued help — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

[2805413893]

The current test image using your code is 160*180. I would like to ask about the size of the test indicator in your paper?

[Yshouang]

There is a big difference between the indexes I tested, MSE = 0.0102 ... SSIM = 0.883 ... PSNR = 24.810 ... which may be better. At 2021-12-03 11:29:24, "candy" ***@***.***> wrote: The current test image using your code is 160*180. I would like to ask about the size of the test indicator in your paper? — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

[sprout81]

您好，请问作者使用的数据集您还有吗，如果有的话，可以发我一份吗，谢谢您