search

anishathalye / neural-style

Neural style in TensorFlow! 🎨
https://anishathalye.com/an-ai-that-can-mimic-any-artist/
GNU General Public License v3.0
5.54k stars 1.52k forks source link

Question: Are the parameters of VGG updated during image synthesis? #88

Closed TheIllusion closed 7 years ago

TheIllusion commented 7 years ago

I've understand that the synthesized output image is generated by minimizing style and content loss.

During iterations, are the parameters of VGG also updated?

As far as I understand, only the input noise image is the only parameter that will be updated. And the parameters of VGG is fixed during the whole process.

Is my understanding right?

Thanks for your great work.

anishathalye commented 7 years ago

Correct, the only trainable variable is the input image (that can be initialized as random noise).

On Jun 27, 2017, at 11:02 PM, TheIllusion notifications@github.com wrote:

I've understand that the synthesized output image is generated by minimizing style and content loss.

During iterations, are the parameters of VGG also updated?

As far as I understand, only the input noise image is the only parameter that will be updated. And the parameters of VGG is fixed during the whole process.

Is my understanding right?

Thanks for your great work.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

TheIllusion commented 7 years ago

Great. Now everything is clear in my mind. Thank you.