Waifu2x

This article covers installing, using and training waifu2x, image super-resolution for anime-style art using deep convolutional neural networks.

Installation

To directly use waifu2x, install waifu2x-git^AUR package. There are other alternates for using waifu2x, just search waifu2x in AUR.

Tip: If you have an NVIDIA GPU, you can install cuda to significantly speed up the conversion process.

Usage

waifu2x is avaliable with command waifu2x. For detailed options, run waifu2x --help

Upscaling

Use --scale_ratio parameter to specify scale ratio you want. And -i with input file name, -o with output file name:

waifu2x --scale_ratio 2 -i my_waifu.png -o 2x_my_waifu.png

Noise Reduction

Use --noise_level parameter(1 or 2) to specify noise reduction level:

waifu2x --noise_level 1 -i my_waifu.png -o lucid_my_waifu.png

And you can use --jobs to specify number of threads launching at same time, benifit for multi-core CPU :

waifu2x --jobs 4 --noise_level 1 -i my_waifu.png -o lucid_my_waifu.png

Upscaling & Noise Reduction

--scale_ratio and --noise_level can be combined, so you can:

waifu2x --scale_ratio 2 --noise_level 1 -i my_waifu.png -o 2x_lucid_my_waifu.png

Tip: If you are finding a batch operation interface, you can have a look at this waifu2x wrapper script

Training

To train custom models, an NVIDIA graphical card is required because waifu2x uses CUDA for computing. Then you need to prepare below develop dependencies and waifu2x source.

Dependencies

Install:

• lua51
• cuda
• snappy
• graphicsmagick
• torch7-git^AUR
• torch7-trepl-git^AUR
• torch7-sys-git^AUR
• torch7-cutorch-git^AUR
• torch7-nn-git^AUR
• torch7-cunn-git^AUR
• torch7-image-git^AUR
• torch7-xlua-git^AUR
• torch7-dok-git^AUR
• torch7-optim-git^AUR
• lua51-graphicsmagick-git^AUR
• lua51-cjson^AUR
• lua51-csvigo-git^AUR
• lua51-snappy-git^AUR

It is recommended to install below optional cuDNN library and bindings package. With them you can enable cuDNN backend for training, which have a significant speed up.

You need to manually download a cudnn binary pack from NVIDIA cuDNN site during installing cudnn.

• (optional)cudnn
• (optional)torch7-cudnn-git^AUR:

waifu2x source

Fetch waifu2x source code from GitHub:

git clone --depth 1 https://github.com/nagadomi/waifu2x.git

Enter source directory. Now you can test waifu2x command line tool:

th waifu2x.lua

Command line tools

Note: If you have installed cuDNN library, you can use cuDNN with -force_cudnn 1 option. cuDNN is too much faster than default kernel.

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 0 -i input_image.png -o output_image.png
th waifu2x.lua -m noise -noise_level 2 -i input_image.png -o output_image.png
th waifu2x.lua -m noise -noise_level 3 -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 0 -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
th waifu2x.lua -m noise_scale -noise_level 3 -i input_image.png -o output_image.png

For more, see waifu2x#command-line-tools.

Train your own models

Note: If you have installed cuDNN library, you can use cuDNN kernel with -backend cudnn option. And, you can convert trained cudnn model to cunn model with tools/rebuild.lua.

Note: The command that was used to train for waifu2x's pretraind models is available at appendix/train_upconv_7_art.sh, appendix/train_upconv_7_photo.sh. Maybe it is helpful.

Data Preparation

Genrating a file list.

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

Note: You should use noise free images.

Converting training data:

th convert_data.lua

Train a Noise Reduction(level1) model

mkdir models/my_model
th train.lua -model_dir models/my_model -method noise -noise_level 1 -test images/miku_noisy.png
# usage
th waifu2x.lua -model_dir models/my_model -m noise -noise_level 1 -i images/miku_noisy.png -o output.png

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

Train a Noise Reduction(level2) model

th train.lua -model_dir models/my_model -method noise -noise_level 2 -test images/miku_noisy.png
# usage
th waifu2x.lua -model_dir models/my_model -m noise -noise_level 2 -i images/miku_noisy.png -o output.png

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

Train a 2x UpScaling model

th train.lua -model upconv_7 -model_dir models/my_model -method scale -scale 2 -test images/miku_small.png
# usage
th waifu2x.lua -model_dir models/my_model -m scale -scale 2 -i images/miku_small.png -o output.png

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

Train a 2x and noise reduction fusion model

th train.lua -model upconv_7 -model_dir models/my_model -method noise_scale -scale 2 -noise_level 1 -test images/miku_small.png
# usage
th waifu2x.lua -model_dir models/my_model -m noise_scale -scale 2 -noise_level 1 -i images/miku_small.png -o output.png

You can check the performance of model with models/my_model/noise1_scale2.0x_best.png.

For latest information, see waifu2x#train-your-own-model.

Docker

See waifu2x#docker.

See also

• waifu2x GitHub repository