ChainerSPIRAL

A modified implementation of Synthesizing Programs for Images using Reinforced Adversarial Learning (SPIRAL) using ChainerRL and MyPaint.

Dependencies

• Pipenv
• MyPaint
• Docker

Run pre-trained models on Docker

cd docker
docker build . -t chainer_spiral
docker run -t --name run_chainer_spiral_demo chainer_spiral pipenv run python demo.py movie trained_models/quickdraw/68976000 result.mp4 --without_dataset
docker cp run_chainer_spiral_demo:/chainer_spiral/ChainerSPIRAL/result.mp4 .

If docker cp ... doesn't work because of a permission error, change permission of the current directory by chmod a+rw .

You can choose a demo mode from static, many, and movie (shown the above):

An example of static demo:

Many demo:

How to setup manually

Install dependencies (CentOS)

sudo yum install gcc gobject-introspection-devel json-c-devel glib2-devel git python autoconf intltool gettext libtool swig python-setuptools gettext gcc-c++ python-devel numpy gtk3-devel pygobject3-devel libpng-devel lcms2-devel json-c-devel gtk3 gobject-introspection

Install libmypaint

git clone https://github.com/mypaint/libmypaint
cd libmypaint
git checkout 0c07191409bd257084d4ea7576deb832aac8868b
./autogen.sh
./configure --prefix=<your-installation-prefix>
make install

Install mypaint-brushes

git clone  https://github.com/mypaint/mypaint-brushes.git
cd mypaint-brushes
git checkout 769ec941054725a195e77d8c55080344e2ab77e4
./autogen.sh
./configure --prefix=<your-installation-prefix>
make install

Build MyPaint with python support

mkdir build_mypaint && cd buid_mypaint
git clone https://github.com/mypaint/mypaint.git
cd mypaint
git checkout 57685af8dbd65719d7874bc501094bade85d94e7
cd ../
pipenv install --python 3.6
pipenv install numpy pygobject pycairo
pipenv shell
cd mypaint
python setup.py build
readlink -f build/lib.linux-x86_64-3.6  # append this path to .env file

Set envrionment variables

Make sure that <your-installation-prefix>/lib is in LD_LIBRARY_PATH and PYTHONPATH. Also PKG_CONFIG_PATH shoud include <your-installation-prefix>/lib and <your-installation-prefix>/share.

Install this project's dependencies

pipenv run install

Check your installation

Go to this repo's directory and run tests by pipenv run test

Train model from scrach

pipenv run python train.py settings/default.yaml logs

Details of training options are available on comments of settings/default.yaml.