search

AmeryXiong / MixDehazeNet

Code for "MixDehazeNet: Mix Structure block for image dehazing network"
MIT License
59 stars 7 forks source link

problems with pretrained model on ITS dataset #6

Open cxl-ustb opened 11 months ago

cxl-ustb commented 11 months ago

Hello,thank you for your good work.I have a problem,when I test the Indoor pretrain model on ITS train set,its psnr is around 22.But when I test the Indoor pretrain model on SOTS test set,its psnr is around 43.Why?

AmeryXiong commented 11 months ago

Because of different test datasets, the test results are different.

cxl-ustb commented 11 months ago

Because of different test datasets, the te Thank you.which dataset you used when you train the pretrain model MixDehazeNet-l?not RESIDE ITS?

AmeryXiong commented 11 months ago

I train the model in RESIDE-ITS dataset and test it in SOTS-Indoor test dataset. You can see the details in my paper.

cxl-ustb commented 11 months ago

I train the model in RESIDE-ITS dataset and test it in SOTS-Indoor test dataset. You can see the details in my paper.

Thank you very much for your reply.I have seen this before. I would like to borrow your training results, which are the results of the model on the training set. But I used your pre trained model to infer the results on the entire training set, and the psnr is only a little over 20. Should the PSNR of your training on the training set not be greater than the results on the test set?

AmeryXiong commented 11 months ago

我没有关注过训练集中推理的结果，我认为可能是因为训练集中图片数量过大，导致训练集推理结果不太好。 或许其他模型在训练集中的推理结果也不太好。

获取Outlook for Androidhttps://aka.ms/AAb9ysg

From: cxl-bupt @.> Sent: Sunday, September 24, 2023 2:43:52 PM To: AmeryXiong/MixDehazeNet @.> Cc: AmeryXiong @.>; Comment @.> Subject: Re: [AmeryXiong/MixDehazeNet] problems with pretrained model on ITS dataset (Issue #6)

I train the model in RESIDE-ITS dataset and test it in SOTS-Indoor test dataset. You can see the details in my paper.

Thank you very much for your reply.I have seen this before. I would like to borrow your training results, which are the results of the model on the training set. But I used your pre trained model to infer the results on the entire training set, and the psnr is only a little over 20. Should the PSNR of your training on the training set not be greater than the results on the test set?

― Reply to this email directly, view it on GitHubhttps://github.com/AmeryXiong/MixDehazeNet/issues/6#issuecomment-1732502498, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AYNYMOC67O347PJCT6AT27DX37JCRANCNFSM6AAAAAA5ETIPUI. You are receiving this because you commented.Message ID: @.***>