Face Super-resolution Demo Based on SR3

Brief

This is a demo for face super-resolution based on SR3 model. It is built by gradio.

Usage

You can simply run the demo as following steps:

1. Install environment

pip install requirements.txt

2. Download pretrain model

We randomly sampled 20,000 pictures from a total of 70,000 pictures from FFHQ dataset for model training, and the iteration is 1,000,000. You can download our pretrain model via the link below:
Google Drive
After the download is complete, put it in the experiments folder.

3. Run app.py

Demo Screenshot

Acknowledge

We are benefiting a lot from the following project:
Image Super-Resolution via Iterative Refinement