Artistic Style Transfer Android Demo

Demonstrates how to setup TensorFlow for artistic style transfer

Android Demo App

Building the app should be straight-forward by opening the project in Android Studio or running ./gradlew assembleDebug on the command line. Android SDK must be installed properly.

Training a style

tbw

Links

Artistic Style Transfer

• Original Paper: "A Neural Algorithm of Artistic Style", Leon A. Gatys, Alexander S. Ecker, Matthias Bethge, 2015
• What Does A.I. Have To Do With This Selfie? - YouTube video explaining the theory behind
• Google Magenta Multistyle Pastiche Generator

Tensorflow

• Tensorflow Mobile (Anroid, iOS)
• Installing Tensorflow
• Tensorflow for poets 1 - Great introduction on how to use tensorflow on mobile and how to optimize models for mobile use.
• Tensorflow for poets 2 - More mobile application tipps & tricks

Implementations

• Logan Lengström's fast style transfer This is the one I ended up using since setting up the training was easy and the results great. It was also used to create the style.pb in this demo.
• Google's Magenta Model (allows mixing of styles): GitHub, blog post and Google Code Lab
• neural-style has an implementation that allows tweaking of the result by specifying more parameters for the training process, like color-preserving or chaning the level of how abstract the result gets

Other

• Google Lab: Optimize models for mobile