Neural Style Transfer with AdaIN

Neural Style Transfer as proposed by Gatys et. al. was a slow and iterative method that could not transfer style in real time. With Adaptive Instance Normalization we achieve arbitrary style transfer in real time.

Our (with @Ritwik_Raha) Keras blog post on AdaIN got published.
Link: Neural Style Transfer with AdaIN

The post is also accompanyed by a hugging face demo for you all to try the models out.
Hugging face demo: https://huggingface.co/spaces/ariG23498/nst

Hope this example turns out to be a good addition to the community.

Results

Nice work Ari!!

Why don’t you publish these models also on TensorFlow Hub?

1 Like

Thanks @lgusm for your kind words and suggestion.

Due to the fact that this was an example I did not train the model with much data to give me high quality results (as one might notice in the hugging face demo).

I will definitely contribute to the hub once I have a better model to share. Thanks again :hugs:

1 Like