A problem when reproduced HyperTransformer code

[fyoung26]

Hello Chaminda,

Thank you for the code and work on HyperTransformer. When I tried to reproduce the score on the botswana4 dataset, the metrics I got are far from expectation:

pretrain: {"loss": 0.07473030593246222, "cc": 0.9290400743484497, "sam": 3.0020320415496826, "rmse": 0.021659649908542633, "ergas": 0.6483104825019836, "psnr": 27.892120361328125}

train： {"loss": 0.08555552270263433, "cc": 0.8993576765060425, "sam": 3.35520076751709, "rmse": 0.02654602937400341, "ergas": 0.7366535663604736, "psnr": 26.585630416870117}

I also used the trained model you provided, and I got: {"loss": 0.05360260047018528, "cc": 0.9539724588394165, "sam": 2.2932522296905518, "rmse": 0.01636636257171631, "ergas": 1.8692368268966675, "psnr": 30.393962860107422}

Both results are far from expectation.

Then,I checked the github issue of HyperTransformer, then I changed "max_value": 8000 to "max_value": 9816(I got this value from the proccess code of matlab),the pretrain metrics got improvement:

{"loss": 0.03306201007217169, "cc": 0.964657187461853, "sam": 1.863145351409912, "rmse": 0.01357241254299879, "ergas": 0.3927982747554779, "psnr": 32.014068603515625}

But still far away from expectation.

Do you know how to solve this problem?

[fyoung26]

I tried to contact you via email(wbandar1@jhu.edu), but dont get reply, I really need your help, thank you.T T

[wgcban]

@fyoung26 I am not exactly sure what is the reason for lower scores on the Botswana dataset. Let me look into it.

[fyoung26]

@wgcban Thank you sincerely, I'm looking for your reply💖

[maozan]

Hi @fyoung26 . May I ask, did you solve your puzzle?

[fyoung26]

Hi @fyoung26 . May I ask, did you solve your puzzle?

Hi @maozan , I didn't solve it yet, do you have any idea about?

[HaiMaoShiTang]

@fyoung26我不确定博茨瓦纳数据集得分较低的原因是什么。让我调查一下。

Hello Sir, Thank you for the code and work on HyperTransformer. I am very confused about the data set processed by matlab. Could you please send me a copy of the processed data set? Email: xiaozhengwang95@gmail.com. Thank you very much.

[HaiMaoShiTang]

@wgcban Hello Sir, Thank you for the code and work on HyperTransformer. I am very confused about the data set processed by matlab. Could you please send me a copy of the processed data set? Email: xiaozhengwang95@gmail.com . Thank you very much.

[wgcban]

@HaiMaoShiTang Hi, I am not sure about I still have the dataset saved in my machine since I am not continuing right now, but I will check and respond to you tomorrow. Thanks.

[HaiMaoShiTang]

Hello, thank you very much for your reply. It will help me a lot. Thank you very much.

—————————————————————————————— Xiaozheng Wang    Machine Learning/ Remote Sensing Image Fusion School of Control Science and Engineering,  Tiangong University. Tianjin,  China. E-mail: @.***

发件人: Chaminda Bandara 发送时间: 2023年3月12日 21:53 收件人: wgcban/HyperTransformer 抄送: HaiMaoShiTang; Mention 主题: Re: [wgcban/HyperTransformer] A problem when reproducedHyperTransformer code (Issue #8)

@HaiMaoShiTang Hi, I am not sure about I still have the dataset saved in my machine since I am not continuing right now, but I will check and respond to you tomorrow. Thanks. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

[hachreak]

Hi @wgcban , it looks like there are many issues with the code and also with the pretrained.

E.g. dataset padova:

I downloaded your pretrained of N_modules 8 for pavia and I tried to compute the PSNR, running the test function.

The test function returns a PSNR of 41.60820770263672 but in the paper you wrote 43.80!

How can I replicate your results? I'm doing something wrong? Could you please share the your datasets? Thanks a lot for your help!

[ChenJFU]

@HaiMaoShiTang Hi, I am not sure about I still have the dataset saved in my machine since I am not continuing right now, but I will check and respond to you tomorrow. Thanks.

Hello Sir, Thank you for the code and work on HyperTransformer. I'm also confused about the preprocessed data set. Could you please send me a copy of the processed data set? My Email: chenjiafu113@gmail.com. Thank you very much.