In this demo, we present a very fast CNN-based style transfer system running on normal iPhones. The proposed app can transfer multiple pre-trained styles to the video stream captured from the builtin camera of an iPhone around 140ms (7fps). We extended the network proposed as a real-time neural style transfer network by Johnson et al. [1] so that the network can learn multiple styles at the same time. In addition, we modified the CNN network so that the amount of computation is reduced one tenth compared to the original network. The very fast mobile implementation of the app are based on our paper [2] which describes several new ideas to implement CNN on mobile devices efficiently. Figure 1 shows an example usage of DeepStyleCam which is running on an iPhone SE.
- Youtube link is here
- Multi Style Transfer Mobile Implementation with Chainer running on iPad Pro 12.9
- Youtube link is here
- DeepStyleCam: A Real-time Multi-Style Transfer App on iOS
- Youtube link is here
- The Prototype
- C for Neural Network Engine
- Objective-C for iOS Programming
- OpenCV
- iOS >= 11.0
- Xcode >= 9.0
git clone https://github.com/negi111111/DeepStyleCam.git
cd ./DeepStyleCam
- Compile project with Xcode
If you use this app in a publication, a link to or citation of this repository would be appreciated.
@InProceedings{tann17,
author="Tanno, R. and Yanai, K.",
title="DeepStyleCam: A Real-time Style Transfer App on iOS",
booktitle="Proc. of International MultiMedia Modeing Conference (MMM)",
year="2017"
}
MIT. Copyright (c) 2017 Ryosuke Tanno