You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Dear all,
Does anybody know the real difference of the model(net) written in the code between this repository and Johnson's fast-neural-style? I mean, in this repo, the default model parameter is 'johnson', If I use the 'johnson' mode, and since the Johnson's newest version has implement 'InstanceNormalization' yet, What is the real difference? and the output is different too. Anybody knows? I'll appreciate any reply.
this repo's output:
Johnson's papers output:
The text was updated successfully, but these errors were encountered:
Dear all,
Does anybody know the real difference of the model(net) written in the code between this repository and Johnson's fast-neural-style? I mean, in this repo, the default model parameter is 'johnson', If I use the 'johnson' mode, and since the Johnson's newest version has implement 'InstanceNormalization' yet, What is the real difference? and the output is different too. Anybody knows? I'll appreciate any reply.
this repo's output:
Johnson's papers output:
The text was updated successfully, but these errors were encountered: