Parameters used for training

[iegorval]

Good afternoon,

May I ask if what were the parameters used for the training in the final version of the model, whose results were shown in the article? Are those all the same as the defaults from the code + flags from the https://github.com/silviazuffi/smalst/blob/master/scripts/smalst_train_run.sh? Also, are the parameters from the first (commented) line of the script are used, or from the second line (in that case, what is nokap)?

Additionally, may I ask what is the difference between flags "is_optimization" and "is_refinement"? The have the same comment in the flag definition, but they seem to mean different things in the code.

I have rewritten the code a bit to use Python 3 with the corresponding new Nvidia drivers, but I am now not sure that I would get the equivalent results due to the big number of parameters. (In case I am able to reproduce it, I would be also happy to share somehow the version for Python 3, so that more people would be able to use it with newer libraries - if you want and agree, of course.)

Thank you a lot in advance for any answer.

Best regards,
Valeria

[silviazuffi]

Hi Valeria,

May I ask if what were the parameters used for the training in the final version of the model, whose results were shown in the article? Are those all the same as the defaults from the code + flags from the https://github.com/silviazuffi/smalst/blob/master/scripts/smalst_train_run.sh <https://github.com/silviazuffi/smalst/blob/master/scripts/smalst_train_run.sh>? Also, are the parameters from the first (commented) line of the script are used, or from the second line (in that case, what is nokap)?

The parameters are those in the code, the commented line was a test (only 40 epochs), the one used for the paper is the one that is not commented. Nokap is a directory where I placed the training data, you can use any name.

Additionally, may I ask what is the difference between flags "is_optimization" and "is_refinement"? The have the same comment in the flag definition, but they seem to mean different things in the code.

is_refinement is something I tried for refining the network weights, but it is not used, is_optimization is explained in the paper, and refers to the optiization step that we perform on each test images.

I have rewritten the code a bit to use Python 3 with the corresponding new Nvidia drivers, but I am now not sure that I would get the equivalent results due to the big number of parameters. (In case I am able to reproduce it, I would be also happy to share somehow the version for Python 3, so that more people would be able to use it with newer libraries - if you want and agree, of course.)

Yes, of course, I think it would be very useful. You can contact me in private and we can have a chat if there is something not very clear. Thanks! Silvia

…

Thank you a lot in advance for any answer. Best regards, Valeria — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#20>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABNYVFMOSQRTBHWAGLBCU7TTG4Q7DANCNFSM42KJNAHQ>.