NeuralStyle Plugin (日本語)

This plugin effect transfers style of style image to content image using neural network.

The model files of neural networks are not contained in this repository. You can get them from nin_imagenet.caffemodel and VGG_ILSVRC_16_layers.caffemodel.

How to install

OS X

TBD

Windows

w/ GPU

0. install Visual Studio 2013

• link1

0. install CUDA Toolkit 7.5 (Windows x86_64)

• link2

0. install cuDNN v4 Library for Windows

• link3
• copy bin, include, and lib directories to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5.

0. install Python 2.7 WIndows 64-bit

• link4

0. install openopt

• exec pip install openopt on command prompt

0. install cvxopt

• download cvxopt-1.1.7+openblas-cp27-none-win_amd64.whl from link5
• exec pip install cvxopt-1.1.7+openblas-cp27-none-win_amd64.whl on command prompt

0. install chainer

• exec set VS100COMNTOOLS=%VS120COMNTOOLS%` ; set PATH=%VS120COMNTOOLS%\..\..\VC\bin;%PATH%` ; pip install chainer on command prompt

0. copy neural_style_synthesizer directory and DWANGO_NeuralStyle.plugin to ${path-to-stuff}/plugins
1. copy nin_imagenet.caffemodel and VGG_ILSVRC_16_layers.caffemodel to ${path-to-stuff}/plugins/neural_style_synthesizer

w/o GPU

0. install Python 2.7 WIndows 64-bit

• link4

0. install openopt

• exec pip install openopt on command prompt

0. install cvxopt

• download cvxopt-1.1.7+openblas-cp27-none-win_amd64.whl from link5
• exec pip install cvxopt-1.1.7+openblas-cp27-none-win_amd64.whl on command prompt

0. install chainer

• exec pip install chainer on command prompt

0. copy neural_style_synthesizer directory and DWANGO_NeuralStyle.plugin to ${path-to-stuff}/plugins
1. copy nin_imagenet.caffemodel and VGG_ILSVRC_16_layers.caffemodel to ${path-to-stuff}/plugins/neural_style_synthesizer

How to use

input ports

port name
Content	content image
Style	style image

parameters

param name	default value	min value	max value
xsplit	1.000	1	16.0	the number of horizontal style image splits
ysplit	1.000	1	16.0	the number of vertical style image splits
resize	1.000	0	1.0	resize factor: 1 (original size), 0.5 (half size), 0.1 (1/10 size), and so on.
iteration	10.000	0	1000.0	the number of iterations for style optimization
method	1.000	0	1.0	method flag for neural network: {0: max_pooling, 1: average_pooling}. The default value is recommended.
model	0.000	0	1.0	model of neural network: {0: nin, 1: vgg}, nin is a small and vgg is a large model (*)
smooth	0.000	0	1.0	sequential processing flag: {0: off, 1: on} (*)
optimize	1.000	0	1.0	whether to optimize the mixture ratio of style image parts: {0: off, 1: on}. The default value is recommended.
gpu	-1.000	-1	3.0	GPU id (CPUs are used when gpu euqals to -1)
content_weight	0.005	0	0.1	content image weight for style optimization
pause	0.000	0	1.0	do-not-automatically-close-consoles flag for debugging: {0: off, 1: on} (only for Windows)

(*) You have to set Dedicated CPUs to Single at Preview Settings and Output Settings, when you use the vgg model or enable the smooth flag.

The first execution is time-consuming, because the plugin generates a cache of the neural network models.