Real-ESRGAN-Docker

This uses the following repo: https://github.com/xinntao/Real-ESRGAN

GitHub Image references:

Images pushed to dockerhub as well as GitHub

docker pull ralphv/realesrgan

docker pull ralphv/realesrgan-with-models

OR

docker pull ghcr.io/ralphv/realesrgan

docker pull ghcr.io/ralphv/realesrgan-with-models

How to build

make build

By default, this will create three images:

• ralphv/realesrgan-base: Base image functional without volumes
• ralphv/realesrgan: The main image to use that creates some volumes to retain downloads
• ralphv/realesrgan-with-models: Image with pre-downloaded models (recommended)

WARNING: The produced images are very large, close to 10 GB

Example runs:

• Run to get the help information:

  docker run --rm ralphv/realesrgan

• Run on a sample input file test.jpg to produce the output as test_out.jpg in same folder

  docker run --rm --gpus all -v .:/data ralphv/realesrgan-with-models -o /data/ -i /data/test.jpg

• Run on all files in current directory

  docker run --rm --gpus all -v .:/data ralphv/realesrgan-with-models -o /data/ -i /data/

Modify image names:

You can create .env file and define a custom image name using the variable IMAGE_NAME

IMAGE_NAME: <custom>/realesrgan

Useful command line arguments

• --face_enhance: Enhances faces but sometimes can create weird faces.
• --fp32: Use fp32 precision during inference. Default: fp16 (half precision).
• -s 2: The scaling factor, can be put to 1 to just enhance images.

You can find more info at the main repo of the library here.

Useful aliases

alias enhance='docker run --rm --gpus all -v .:/data ralphv/realesrgan-with-models --fp32 -i /data/ -o /data/ -s 1'

alias enlarge='docker run --rm --gpus all -v .:/data ralphv/realesrgan-with-models --fp32 -i /data/ -o /data/ -s 2'

alias enhancef='docker run --rm --gpus all -v .:/data ralphv/realesrgan-with-models --face_enhance --fp32 -i /data/ -o /data/ -s 1'

alias enlargef='docker run --rm --gpus all -v .:/data ralphv/realesrgan-with-models --face_enhance --fp32 -i /data/ -o /data/ -s 2'