GFPGAN-webui

Based on the original GFPGAN, built with gradio, tailored for my use case.

• Simple to use, one-click inference
• Could take advantage of AMD GPU and be able to release the it to save power after each run
• Allow retrieval of previously processed images in case of connection failure
• Easy deployment and replication through Docker image
• Now support GPEN

Web UI screenshot

Development

1. Rename the corresponding pyproject.*.toml file of your target runtime to pyproject.toml

2. Install dependencies with Pypoetry

poetry install --no-root

3. Run with scripts/run_cpu.sh or scripts/run_rocm.sh if an AMD GPU is available.

Note that in order to run with AMD GPUs, ROCm package is required, instruction on the installation could be found over ollama development docs.

Build Docker images

Poetry package manager is required to generate a requirements.txt file for docker build.

Install the export plugin

poetry self add poetry-plugin-export

Lock dependencies

poetry lock

Build with ROCm

./scripts/build_docker_rocm.sh`

Or with CPU-only

./scripts/build_docker_cpu.sh`

Usage

Download all models to ./weights directory

./scripts/download_models.sh

Rename the corresponding docker-compose.*.yaml file of your target runtime to docker-compose.yaml

docker compose up -d

Note that the ROCm image is massive, almost 20GB when compressed and 60GB to store locally. The CPU-only image is available on Docker Hub, but a image build is required to run ROCm version.