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

Loss function #172

Closed junk-zqm closed 3 years ago

junk-zqm commented 3 years ago

Is there something wrong with the loss function?The results are very large： content loss: 164342 style loss: 41753.8 tv loss: 26570.2 total loss: 232666

anishathalye commented 3 years ago

No, that's normal

junk-zqm commented 3 years ago

Is this data compared with zero?The result seems to be terrible.

anishathalye commented 3 years ago

I am not sure what you mean by compared with zero. The goal is to minimize the loss, but for the definition of the loss function in the paper, it is not necessarily possible to achieve zero loss.

junk-zqm commented 3 years ago

I mean how to use this loss to judge the quality of this transfer.

anishathalye commented 3 years ago

I don't think there's a good answer to that question besides doing it manually/qualitatively.