How to choose proper google style transfer model?

49 views Asked by At

please help!

I've started to doing a task using Google steps from there: https://www.tensorflow.org/tutorials/generative/style_transfer

Everything works fine with demo images which is stored by Google, but if i start to do the same job with my own pictures - result is not so relevant.

I tried to open an URL from the code with exact used model and i see this model storage https://tfhub.dev/google

but there is a lot of models. How can i get a proper one?

I tried to brute force through the models but got a lot of not appropriate models which is even not suitable for this type of task.

Please help me with solution - how can i get or what is correct way to search a proper model for this task?

1

There are 1 answers

0
Andrey Khorlin On

Currently for style transfer there is a single model https://tfhub.dev/google/magenta/arbitrary-image-stylization-v1-256/2. There are plenty of generic image model useful for other image-based tasks such as classification. If you train, your own style transfer model, you can share it with the community by publishing it on TensorFlow Hub.