INSTALL-mac.md

0. Limitations

Currently, it works very limitedly on macOS. (2023-05-18)

swin_unet/* models (waifu2x's default models) do not work on MPS backend. Only the older models upcunet, cunet, upconv_7, and vgg_7 will work. Use older models such as CUNet or use CPU backend.

To use CUNet models,

• Specify --model-dir waifu2x/pretrained_models/cunet/art option for waifu2x.cli
• Specify --art-model-dir waifu2x/pretrained_models/cunet/art option for waifu2x.web

To use CPU backend,

• Specify --gpu -1 option for waifu2x.cli
• Specify --gpu -1 option for waifu2x.web

1. Install dependencies packages

Install Python 3.10+ (3.10+ is required for macOS. related to urllib3/urllib3#2168 )

brew install python3

Install ImageMagick and libraqm.

brew install imagemagick libraqm

2. Clone

git clone https://github.com/nagadomi/nunif.git
cd nunif

3. Setup virtualenv (optional)

initialize

python3 -m venv .venv

activate

source .venv/bin/activate

(exit)

deactivate

4. Install PyTorch

pip3 install -r requirements-torch.txt

5. Install pip packages

pip3 install -r requirements.txt

6. Run waifu2x.web

Download pre-trained models.

python -m waifu2x.download_models

Generate waifu2x/web/public_html

python -m waifu2x.web.webgen

Start the web server.

python -m waifu2x.web --art-model-dir ./waifu2x/pretrained_models/cunet/art --photo-model-dir ./waifu2x/pretrained_models/upconv_7/photo

When using CPU backend

python -m waifu2x.web --gpu -1

Open http://localhost:8812/ (style=art_scan does not work)

See also waifu2x README.md.