Reproducing demo models

[yosh7]

Hi, kudos for an awesome project ! I managed to run this using tf 1.13.1 and works perfectly. However, when trying to train I get poor results , no where near the pre-trained models. After trying with the defaults , I also tried changing some of the hyperparameters, eg. content/tv/style weights , learning rate, batch sizes. I see that learn rate and batch sizes have significant effect on end results ,but can't reproduce your models. Can you please publish the parameters used for training ? style image sizes ? any other modifications to the code that can help improve results ?

[dakshvar22]

Hi @yosh7 , were you able to re-produce the pre-trained models?

[yosh7]

Yep. I installed everything from scratch, Linux, cuda, cudnn (this one's important!) using latest packages for everything and got good results. I assume there was a package compatibility issue before, which is strange as the project ran with no errors, just produced bad results. @dakshvar22

[dakshvar22]

Can you tell the version of tensorflow you used?

On Mon, Jul 6, 2020, 13:49 yosh7 [email protected] wrote:

Yep. I installed everything from scratch, Linux, cuda, cudnn (this one's important!) using latest packages for everything and got good results. I assume there was a package compatibility issue before, which is strange as the project ran with no errors, just produced bad results.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/hwalsuklee/tensorflow-fast-style-transfer/issues/21#issuecomment-654185626, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACCOBGIER34NMI5WVPNLTZTR2G237ANCNFSM4M66QBMA .

[yosh7]

TF2. And now that I'm thinking of it you may also need to fix the code imports. Use: Import tensorflow.compat.v1 as tf In all TF imports.

[dakshvar22]

@yosh7 I think just the import won't be enough as I am running into some other errors as well when I upgrade to TF 2.2 .Would you be able to share your updated TF2 code?

[yosh7]

Yes. The problem was with drivers and Cuda installations + tfgpu package I used. I tried over and over several combinations until it finally worked

On Mon, Aug 2, 2021, 4:13 PM guyash @.***> wrote:

@yosh7 https://github.com/yosh7 did you manage to re-produce the pre-trained models? I train the model on the given style images and get poor results also.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/hwalsuklee/tensorflow-fast-style-transfer/issues/21#issuecomment-891015807, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJWZCGRWVGPV2NSXVHJP5QDT22KWLANCNFSM4M66QBMA .

[guyash]

@yosh7 Does it mean you got the same outputs when running the pre-trained models and the models you trained by yourself on the given style images? If so, can you please tell me what Cuda & tfgpu packages you used? Also, I understand you used the default parameters for training, right?

Thank you so much!