Example of Neural Style Transfer [Johnson+, 2016] with WebDNN

You can download pre-trained model from

• yusuketomoto/chainer-fast-neuralstyle
• gafr/chainer-fast-neuralstyle-models

1. Install WebDNN (see document)

2. Clone gafr/chainer-fast-neuralstyle-models at root directory of this repository. Note: Download of about 200MB model is needed.

   $ git clone https://github.com/gafr/chainer-fast-neuralstyle-models

3. Convert model

   $ python convert.py --model kanagawa --backend webgpu

   --model option selects the style. For details, see python convert.py -h

4. Access to index.html with your browser

This example use Placeholder feature in WebDNN, which supports specifying tensor shape at runtime. However, only WebGPU and WebAssembly backend supports this feature and WebGL backend does NOT support.

MIT License, 2017, Yuichiro Kikura[email protected]