[go: up one dir, main page]

Skip to content

pfnet/PaintsChainer

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Paints Chainer

Paints Chainer is a line drawing colorizer using chainer. Using CNN, you can colorize your sketch semi-automatically .

image

DEMO

http://paintschainer.preferred.tech/

Requirement

If not specified, versions are assumed to be recent LTS version.

  • A Nvidia graphic card supporting cuDNN i.e. compute capability >= 3.0 (See https://developer.nvidia.com/cuda-gpus)
  • Linux: gcc/ g++ 4.8
  • Windows: "Microsoft Visual C++ Build Tools 2015" (NOT "Microsoft Visual Studio Community 2015")
  • Python 3 (3.5 recommended) ( Python 2.7 needs modifying web host (at least) )
  • Numpy
  • openCV "cv2" (Python 3 support possible, see installation guide)
  • Chainer 2.0.0 or later
  • CUDA / cuDNN (If you use GPU)

Installation Guide

check wiki page https://github.com/pfnet/PaintsChainer/wiki/Installation-Guide

Starting web host

UI is html based. using wPaint.js Server side is very basic python server

boot local server python server.py

access to localhost localhost:8000/

Learning

main code of colorization is in cgi-bin/paint_x2_unet

to train 1st layer using GPU 0 python train_128.py -g 0 to train 2nd layer using GPU 0 python train_x2.py -g 0

License

Source code : MIT License

Pre-trained Model : All Rights Reserved

Pre-Trained Models

Download following model files to cgi-bin/paint_x2_unet/models/

http://paintschainer.preferred.tech/downloads/

(Copyright 2017 Taizan Yonetsuji All Rights Reserved.)

Developer Community

Feel free to request an invitation!

https://paintschainerdev.slack.com/

Acknowledgements

This project is powered by Preferred Networks.

Thanks a lot for rezoolab, mattya, okuta, ofk . This project could not be achived without their great support.

Line drawing of top image is by ioiori18.