waifu2x

Image Super-Resolution for anime/fan-art using Deep Convolutional Neural Networks.

Demo-Application can be found at http://waifu2x.udp.jp/ .

Summary

Click to see the slide show.

References

waifu2x is inspired by SRCNN [1]. 2D character picture (HatsuneMiku) is licensed under CC BY-NC by piapro [2].

• [1] Chao Dong, Chen Change Loy, Kaiming He, Xiaoou Tang, "Image Super-Resolution Using Deep Convolutional Networks", http://arxiv.org/abs/1501.00092
• [2] "For Creators", http://piapro.net/en_for_creators.html

Public AMI

AMI name: waifu2x server
AMI ID: ami-75f01931
Region: N. California
Instance: g2.2xlarge (require a GPU)
OS: Ubuntu 14.04
User: ubuntu

Dependencies

Hardware

• NVIDIA GPU (Compute Capability 3.0 or later)

Platform

• Torch7
• NVIDIA CUDA
• NVIDIA cuDNN

Packages (luarocks)

• cutorch
• cunn
• cudnn
• graphicsmagick
• turbo
• md5
• uuid

NOTE: Turbo 1.1.3 has bug in file uploading. Please install from the master branch on github.

Installation

Setting Up the Command Line Tool Environment

(on Ubuntu 14.04)

Install Torch7

sudo apt-get install curl
curl -s https://raw.githubusercontent.com/torch/ezinstall/master/install-all | sudo bash 

see Torch (easy) install

Install CUDA and cuDNN.

Google! Search keyword is "install cuda ubuntu" and "install cudnn ubuntu"

Install packages

sudo luarocks install cutorch
sudo luarocks install cunn
sudo luarocks install cudnn
sudo luarocks install graphicsmagick

Test the waifu2x command line tool.

th waifu2x.lua

Setting Up the Web Application Environment (if you needed)

Install luajit 2.0.4

curl http://luajit.org/download/LuaJIT-2.0.4.tar.gz
tar -xzvf LuaJIT-2.0.4.tar.gz
cd LuaJIT-2.0.4
make
sudo make install

Install packages

Install luarocks packages.

sudo luarocks install md5
sudo luarocks install uuid

Install turbo.

git clone https://github.com/kernelsauce/turbo.git
cd turbo
sudo luarocks make rockspecs/turbo-dev-1.rockspec 

Web Application

Please edit the first line in web.lua.

local ROOT = '/path/to/waifu2x/dir'

Run.

th web.lua

View at: http://localhost:8812/

Command line tools

Noise Reduction

th waifu2x.lua -m noise -noise_level 1 -i input_image.png -o output_image.png

th waifu2x.lua -m noise -noise_level 2 -i input_image.png -o output_image.png

2x Upscaling

th waifu2x.lua -m scale -i input_image.png -o output_image.png

Noise Reduction + 2x Upscaling

th waifu2x.lua -m noise_scale -noise_level 1 -i input_image.png -o output_image.png

th waifu2x.lua -m noise_scale -noise_level 2 -i input_image.png -o output_image.png

See also images/gen.sh.

Training Your Own Model

Data Preparation

Genrating a file list.

find /path/to/image/dir -name "*.png" > data/image_list.txt

(You should use PNG! In my case, waifu2x is trained by 3000 PNG images.)

Converting training data.

th convert_data.lua

Training a Noise Reduction(level1) model

th train.lua -method noise -noise_level 1 -test images/miku_noise.png
th cleanup_model.lua -model models/noise1_model.t7 -oformat ascii

You can check the performance of model with models/noise1_best.png.

Training a Noise Reduction(level2) model

th train.lua -method noise -noise_level 2 -test images/miku_noise.png
th cleanup_model.lua -model models/noise2_model.t7 -oformat ascii

You can check the performance of model with models/noise2_best.png.

Training a 2x UPscaling model

th train.lua -method scale -scale 2 -test images/miku_small.png
th cleanup_model.lua -model models/scale2.0x_model.t7 -oformat ascii

You can check the performance of model with models/scale2.0x_best.png.