Arbitrary Style Transfer with Style-Attentional Networks

eridgd / WCT-TF

TensorFlow/Keras implementation of "Universal Style Transfer via Feature Transforms" from https://arxiv.org/abs/1705.08086

MIT License

319 stars 79 forks source link

Arbitrary Style Transfer with Style-Attentional Networks #26

Closed GlebSBrykin closed 5 years ago

GlebSBrykin commented 5 years ago

How about implementing this algorithm: https://arxiv.org/abs/1812.02342v3? It works faster, but the quality is much higher than what is represented in the WCT.

GlebSBrykin commented 5 years ago

I have an exclusive example obtained directly from the developers. It can be used for quality evaluation. Result from SANet: gogh_transfer Original: content Result from WCT: starrynight_result

o0t1ng0o commented 5 years ago

Hi @GlebBrykin, how did you get the source code of SANet? Or did you implement it yourself?

GlebSBrykin commented 5 years ago

My implementations here: https://github.com/ColorfulSoft/StyleTransfer-Colorization-SuperResolution/tree/master/Style%20Transfer/2018.%20Arbitrary%20Style%20Transfer%20with%20Style-Attentional%20Networks and here: https://github.com/GlebBrykin/SANET. It will be finalized in the near future, but there is another implementation: https://github.com/mumair5393/Style-Transfer-with-Style-Attentional-Networks it looks more complete.