how should I preprocess the data?

[GanZhan]

Hello, I read your paper, it looks like we need to resize the input data into 192×192×60, but the model(EncodeTrans_V3DUNet) seems only accept the 5-dimensional input, so I am kind of confused?
(1) EncodeTrans_V3DUNet is your proposed method SpecTr, right?
(2) what should I preprocess the input hyperspectral image to obtain (b, c, s, w, h), especially the meaning of s?
Could you kindly share your ideas?

[boxiangyun]

Thank you for your attention to my work. Just as most of 3D medical data (such as CT or MRI), we assume that the feature dimension of the input is one (B, C =1, S, H, W).  btw, different from other 3D segmentation tasks, output of spectr model is 2D mask.  We are working on the new version of SpecTr and will be happy to forward it to you in the future.

…

------------------ 原始邮件 ------------------ 发件人: "hfut-xc-yun/SpecTr" ***@***.***>; 发送时间: 2021年10月3日(星期天) 中午11:05 ***@***.***>; ***@***.***>; 主题: [hfut-xc-yun/SpecTr] how should I preprocess the data? (#2) Hello, I read your paper, it looks like we need to resize the input data into 192×192×60, but the model(EncodeTrans_V3DUNet) seems only accept the 5-dimensional input, so I am kind of confused? (1) EncodeTrans_V3DUNet is your proposed method SpecTr, right? (2) what should I preprocess the input hyperspectral image to obtain (b, c, s, w, h), especially the meaning of s? Could you kindly share your ideas? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

[boxiangyun]

We do not have the mask corresponding to RGB images, and due to the reasons of the instrument, we cannot get the strictly corresponding RGB and hyperspectral images. Generally, we can use the corresponding band range of RGB to synthesize pseudo-color images to approximately replace RGB.

…

------------------ 原始邮件 ------------------ 发件人: ***@***.***>; 发送时间: 2021年10月7日(星期四) 中午11:25 收件人: ***@***.***>; 抄送: ***@***.***>; ***@***.***>; 主题: Re: [hfut-xc-yun/SpecTr] how should I preprocess the data? (#2) Thanks for your comment, now I can run your specTr on the dataset. And I met another problem about the data, you see, it seems like the dataset website provides the hyperspectral data and RGB data, I can generate the hyperspectral masks based on the .xml files, but how can I obtain the RGB masks? (I tried to flip the content in RGB, and resize the hyperspectral mask to match the RGB), but it seems not quite fit the RGB content, could you kindly give me some advice on it? — 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.

[GanZhan]

Hello, Thanks for your reply. I am trying to reproduce your result, but it seems like you only use part of the data in your given dataset, I have already downloaded the whole dataset, could you kindly give me the 401 training filenames and 113 testing filenames of your dataset? I can construct the dataset by myself, I will be really thankful if you could help me with this. Hoping this won't cause you much trouble.

[luckyhug]

Also wanted to test on the full code files, I have downloaded the whole dataset on their website @hfut-xc-yun