Training 3-channel input data

[deepshwang]

Your work has been of great help for my research, so thank you very much first of all.

I was trying to implement your work on rotated mnist and CIFAR-10.

Here are my questions..

1)When trying to train CIFAR-10, I tried to leverage your model TIgraNet_eth80 in models.py.

It seems like there's a problem directly using this on 3 channel input with filter_size_in set as 1.

It didn't work of course, so I changed the filter_size_in to be 3 for spectral_conv1. Then, the model training proceeded but the model was barely trained.

Did you grayscale-ized the color image or is there something I am missing???

Thank you!

[RenataKh]

Dear Sungwon,

Thank you for your interest in our work.

Yes, we applied our approach to gray scale images. Regarding the code, please reach out directly to Joel by this email: j oel.fonseca@epfl.ch joel.fonseca@epfl.ch

Best regards, Renata

чт, 24 дек. 2020 г. в 06:12, Sungwon Hwang notifications@github.com:

Your work has been of great help for my research, so thank you very much first of all.

I was trying to implement your work on rotated mnist and CIFAR-10.

Here are my questions..

1)When trying to train CIFAR-10, I tried to leverage your model TIgraNet_eth80 in models.py.

It seems like there's a problem directly using this on 3 channel input with filter_size_in set as 1.

It didn't work of course, so I changed the filter_size_in to be 3 for spectral_conv1. Then, the model training proceeded but the model was barely trained.

Did you grayscale-ized the color image or is there something I am missing???

Thank you!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/LTS4/TIGraNet/issues/1, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAIESBAISKE4GSQE7KYLIUTSWLETVANCNFSM4VHZJ6IA .

[farukuslu]

Hello ReataKh,

Your work is really helpful for my research. I am master's student in Duke University, USA. I am writing to you to ask the data set of TIGraNET. I am using your algorithm in the term project of my machine learning class. I would be so grateful, If you share the link to data set (I assume it is "mnist_012").

With best regards Faruk

[joelmfonseca]

Hi all,

You can download the exact same dataset here. You will be able to recreate all the variations from the dataset.py module.