Graduation project for image restoration, enhancement & colorization.
Run setup.sh
to download weights & models & stuff.
Run main.py
to generate results. Input directory is sample_images
, but you can change that (see run
in main.py
).
According to RunMode
and whether colorize
is set, output is either <output-dir>/face_restore
or <output-dir>/quality_enh/restored_image
or <output-dir>/colorization
(also see run
in main.py
).
- Supplant out-of-the-box face restoration in scratched photos with GPEN's face inpainting. Not sure if it'd work though. We'd have to blend in the restored face to the result of running vanilla scratch restoration.