Different loss function scores for generator and discriminator

[KartikeyKansal1]

Hey, is there a way in cyto-dl to obtain separate loss function scores for the discriminator and the generator?

[benjijamorris]

The GAN loss should be split by generator and discriminator (see mlflow screenshot below). Are you thinking for the generator to split out the GAN loss and the reconstruction loss?

[KartikeyKansal1]

Hi @benjijamorris, thanks for your reply and sorry about the delay in getting back. Yes I'm looking for this only. Is there a way to print this in the metrics.csv?

[benjijamorris]

I think the same metrics get logged regardless of whether you use a csv/tensorboard/mlflow/etc. logger. Running a quick test gave me these columns in my logged csv. Is this different from what you're getting? If so, could you share which columns are logged?

[KartikeyKansal1]

Yes I get different log output

[benjijamorris]

Weird! Is this on main running python cyto_dl/train.py experiment=im2im/gan?

[KartikeyKansal1]

This is im2im/labelfree

…

On Wed, Mar 20, 2024 at 4:48 PM benjijamorris ***@***.***> wrote: Weird! Is this on main running python cyto_dl/train.py experiment=im2im/gan? — Reply to this email directly, view it on GitHub <#350 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ANOVHKSSAG5J4YVUXBEBMKLYZH74XAVCNFSM6AAAAABEQTTF5OVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4DQNJYGQYTS> . You are receiving this because you authored the thread.Message ID: ***@***.*** com>

[benjijamorris]

ah! Labelfree doesn't use a GAN, so you won't see any GAN-related losses being logged. I'm going to close this issue, but if you want to use a GAN, the command I sent above should work very similarly to im2im/labelfree.