hantek / deeplearn_hsi Goto Github PK

Hello, I would like to ask, the downloaded pavia dataset is matlab version, can be opened directly in matlab but nothing is displayed, why? I am not very familiar with the code, so thank you.

Hello, as for experiment on Salinas Valley data, the network parameter [310 100] performs bad, how to tune the depth and number of neurons in each layer ?

I can't explain why, but when I used your KSC_SdA.py code (only changed the file roots), my final test error is above 20% and the Kappa index is close to 0.74, which is pretty far from the results in the "Deep Learning-Based Classification of Hyperspectral Data" article. Did you get those results (OA, AA, Kappa) with 3300 pretraining epochs, a pretraining learning rate of 0.6, 400000 training epochs, a training learning rate of 0.05, mini-batches of 100 and a single layer of 20 autoencoders or did you use other parameters?

I got similar results as in the article with pavia_SdA.py.

Thanks.

Traceback (most recent call last):
File "pavia_SdA.py", line 38, in
img = scale_to_unit_interval(data['Pavia'].astype(theano.config.floatX))
KeyError: 'Pavia'

Hello,

I have successfully installed all libraries via anaconda on WSL2 and Windows 10. In using the KSC test dataset, I had no issues running on my systems CPU, but I was unable to initialize communication with my NVIDIA Quadro RTX 4000 GPU through Theanos, which I believe is enabled through pygpu. Any help solving this issue would be greatly appreciated. Thank you!