about code for training

[JCCC0328]

Hi, thanks for the code, i was interested with your outstanding work. But I am having some difficulties in writing training code to train the model, would you consider making the training code public so that I can train my own model? Thanks a lot!

[kosta-pmf]

Hello,
Our training code is in this script train.py. Take a look and see if it is of use.

[JCCC0328]

Hi, I tried to train with your model on a new dataset, but the model barely converges, and the accuracy is only close to 50%, can you provide the full training code please?

[kosta-pmf]

Hello,
Thank you for your inquiry. We are currently working on publishing our dataset and improving our algorithm. It should be done in the next couple of months. We are expecting permission from the Parliament of Montenegro.
Until then, I suggest you try training on your own data and excluding the attacks part of the network for the start, to see how much data the decoder needs to converge. I suppose that the encoder loss is completely converging and not letting the decoder converge.
If you succeed with that, then you can try introducing the attacks.