search

zilongzhong / SSRN

This is a tensorflow and keras based implementation of SSRNs in the IEEE T-GRS paper "Spectral-Spatial Residual Network for Hyperspectral Image Classification: A 3-D Deep Learning Framework".
Other
211 stars 74 forks source link

Not getting the exact results #11

Open TanmDL opened 5 years ago

TanmDL commented 5 years ago

Thank you for sharing the code. I have tested the code with IndianPines dataset. I have trained the model with 200 runs as it is mentioned on readme file. After 200 runs of training, each element in EachClassAccuracy are obtained as follows, but some of them are showing zeros values. [1. 0.99557913 0.99250375 1. 0.99741602 1. 1. 1. 0. 0.99734748 0.97369727 1.

  1. 0.99457995 0. 0. ] i.e. class- 9, 15,16 are not detected at all. The input patch length is 3.
zilongzhong commented 5 years ago

Please check different settings like learning rates and patch length, and check numbers of data for training.

On Wed, 9 Jan 2019 at 23:40, TanmDL notifications@github.com wrote:

Thank you for sharing the code. I have tested the code with IndianPines dataset. I have trained the model with 200 runs as it is mentioned on readme file. After 200 runs of training, each element in EachClassAccuracy are obtained as follows, but some of them are showing zeros values. [1. 0.99557913 0.99250375 1. 0.99741602 1. 1. 1. 0. 0.99734748 0.97369727 1.

1.

   0.99457995    0.         0.        ] i.e. class- 9, 14,15 are not detected at all.

The input patch length is 3.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/zilongzhong/SSRN/issues/11, or mute the thread https://github.com/notifications/unsubscribe-auth/APQzzucqupUI8S-i6EUd1uA4eADi1qNTks5vBsQfgaJpZM4Z4xjt .

TanmDL commented 5 years ago

Thank you for giving the reply. It is working now. Pls give the reply of #10 issue. I have also the similar questions what he has mentioned.

zilongzhong commented 5 years ago

Check the sizes of train, val, test sets. This is a problem of datasets size consistency between calculated size setting and intended setting.

TanmDL commented 5 years ago

thank you Zilong, it was size issues.  On Friday, January 25, 2019, 5:42:07 AM GMT+11, Zilong Zhong notifications@github.com wrote:

Check the sizes of train, val, test sets. This is a problem of datasets size consistency between calculated size setting and intended setting.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

LemonLov commented 5 years ago

Thank you for giving the reply. It is working now. Pls give the reply of #10 issue. I have also the similar questions what he has mentioned.

Could you tell me how you solve your problem? I have the same problem, and I have tried many ways but it still didn't work. Thank you.

zilongzhong commented 5 years ago

Basically, it's an indexing changing problem, because we need to pad the raw HSI with 3 pixels at each edge. The best way I suggest is to print out old and new indexes to make sure they correspond to each other.

On Mon, 22 Apr 2019 at 03:37, Arthur notifications@github.com wrote:

Thank you for giving the reply. It is working now. Pls give the reply of

10 https://github.com/zilongzhong/SSRN/issues/10 issue. I have also

the similar questions what he has mentioned.

Could you tell me how you solve your problem? I have the same problem, and I have tried many ways but it still didn't work. Thank you.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/zilongzhong/SSRN/issues/11#issuecomment-485352394, or mute the thread https://github.com/notifications/unsubscribe-auth/AD2DHTU56HOKDAQTFOSSBXLPRVTNJANCNFSM4GPDDDWQ .