danfenghong / IEEE_TNNLS_EGU-Net

Danfeng Hong, Lianru Gao, Jing Yao, Naoto Yokoya, Jocelyn Chanussot, Uta Heiden, Bing Zhang. Endmember-Guided Unmixing Network (EGU-Net): A General Deep Learning Framework for Self-Supervised Hyperspectral Unmixing, IEEE TNNLS, 2021.
55 stars 13 forks source link

您好,我在代码上测试Jasper ridge数据集,但是结果出入很大,想请教几个问题 #11

Open OkitaSouji-yume opened 2 years ago

OkitaSouji-yume commented 2 years ago

Jasper ridge数据集上使用[Pseudo_endmembers_generation.m]默认参数的情况下提取出的端元束是2000×198,但是之后将端元束进行解混效果有出入,请问[Pseudo_endmembers_generation.m]代码中是否也需要调整参数

danfenghong commented 2 years ago

首先,jasper ridge每个patch提取的端元是4,此外, [Pseudo_endmembers_generation.m]重要的是提取最后的 TrLabel,因此,对应的 M也是要使用 jasper ridge数据的。

OkitaSouji-Genm @.***> 于2022年6月2日周四 00:52写道:

Jasper ridge数据集上使用[Pseudo_endmembers_generation.m]默认参数的情况下提取出的端元束是2000×198,但是之后将端元束进行解混效果有出入,请问[Pseudo_endmembers_generation.m]代码中是否也需要调整参数

— Reply to this email directly, view it on GitHub https://github.com/danfenghong/IEEE_TNNLS_EGU-Net/issues/11, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFL2GZQMS73NW3MFB7DJSXTVM6INHANCNFSM5XR66HFA . You are receiving this because you are subscribed to this thread.Message ID: @.***>

OkitaSouji-yume commented 2 years ago

非常感谢您无私的分享和回复 在代码中 k = 5; EM = []; for i = 1 : (m / k) for j = 1 : (n / k) sub_X_3d = X_3d((i - 1) k + 1 : k i, (j - 1) k + 1 : k j, :); sub_X_2d = hyperConvert2d(sub_X_3d); [sub_EM, ind, ~] = VCA(sub_X_2d, 'Endmembers', 4, 'SNR', 30); EM = [EM, sub_EM]; end end 是否只需要将[sub_EM, ind, ~] = VCA(sub_X_2d, 'Endmembers', 5, 'SNR', 30)中的5改成4就可以了 这样更改之后所提取的TrLabel维度为1600×4

danfenghong commented 2 years ago

还有下面解混使用到的端元M,你需要根据 jasper ridge数据集进行改变。

OkitaSouji-yume @.***> 于2022年6月2日周四 10:57写道:

非常感谢您无私的分享和回复 在代码中 k = 5; EM = []; for i = 1 : (m / k) for j = 1 : (n / k) sub_X_3d = X_3d((i - 1) k + 1 : k i, (j - 1) k + 1 : k j, :); sub_X_2d = hyperConvert2d(sub_X_3d); [sub_EM, ind, ~] = VCA(sub_X_2d, 'Endmembers', 4, 'SNR', 30); EM = [EM, sub_EM]; end end 是否只需要将[sub_EM, ind, ~] = VCA(sub_X_2d, 'Endmembers', 5, 'SNR', 30)中的5改成4就可以了 这样更改之后所提取的TrLabel维度为1600×4

— Reply to this email directly, view it on GitHub https://github.com/danfenghong/IEEE_TNNLS_EGU-Net/issues/11#issuecomment-1144367008, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFL2GZXZ3FUNUY6FT4LY7IDVNAPK3ANCNFSM5XR66HFA . You are receiving this because you commented.Message ID: @.***>

OkitaSouji-yume commented 2 years ago

代码中的k值是否代表k-means聚类算法聚类个数,这个需要根据数据集的不同进行调整吗,例如在Jasper ridge数据集中,该值依然设定为5

danfenghong commented 2 years ago

k代表对图像进行划分patch的尺寸。

OkitaSouji-yume @.***> 于2022年6月2日周四 11:49写道:

代码中的k值是否代表k-means聚类算法聚类个数,这个需要根据数据集的不同进行调整吗,例如在Jasper ridge数据集中,该值依然设定为5

— Reply to this email directly, view it on GitHub https://github.com/danfenghong/IEEE_TNNLS_EGU-Net/issues/11#issuecomment-1144404859, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFL2GZQFS47BU7WYT65XGCTVNAVKVANCNFSM5XR66HFA . You are receiving this because you commented.Message ID: @.***>

atheeraa commented 1 year ago

to change according to the jas

Hello, Can you please share me the parameters you've changed inorder to work with Jasper dataset?