My12123 / Manga-Colorization-FJ

An amazing manga comic colorization project | 漫画 AI 上色

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

An amazing manga colorization project | 漫画AI上色

You can try it in Open In Colab

If Manga-Colorization-FJ is helpful, please help to ⭐ this repo or recommend it to your friends 😊

New features

  • Skip color img just copy/sr to output.
  • Replace the offered "zipfile" weight to old "pt" format to support pytorch version >=1.0
  • Add tile img option for small cuda memory gpu.
  • Add Real-ESRGAN (support tile) for output super-resolution.
  • Support only SR mode; only Color mode; all Color mode.
  • Support Chinese path.
  • Support Sub-folder mode.

Automatic colorization

  1. Download generator. Put 'generator.pt' in ./networks/
wget https://cubeatic.com/index.php/s/PcB4WgBnHXEKJrE/download -O ./networks/generator.pt
  1. Put imgs into "./input/"
  2. To colorize image or folder of images, use the following command:

USE CPU:

$ python inference.py

USE GPU:

$ python inference.py -g

Only SR mode(no color):

$ python inference.py -onlysr

Color all mode(no skip color one):

$ python inference.py -ca

No SR mode(only color):

$ python inference.py -nosr

Sub-dir mode(handle all sub dir):

$ python inference.py -sub
  1. Colorized image saved to "./output/"

Usage of python script

usage: inference.py [-h] [-p PATH] [-op OUTPUTPATH] [-gen GENERATOR]
                    [-sur SURPERPATH] [-ext EXTRACTOR] [-g] [-nd]
                    [-ds DENOISER_SIGMA] [-s SIZE] [-ct COLORTILE]
                    [-st SRTILE] [--tile_pad TILE_PAD] [-nosr] [-ca] [-onlysr]

optional arguments:
  -h, --help            show this help message and exit
  -p PATH, --path PATH  input dir/file
  -op OUTPUTPATH, --outputpath OUTPUTPATH
                        output dir
  -gen GENERATOR, --generator GENERATOR
  -sur SURPERPATH, --surperpath SURPERPATH
  -ext EXTRACTOR, --extractor EXTRACTOR
  -g, --gpu             Use gpu
  -nd, --no_denoise     No denoiser before color
  -ds DENOISER_SIGMA, --denoiser_sigma DENOISER_SIGMA
                        Denoiser_sigma
  -s SIZE, --size SIZE  Color output size
  -ct COLORTILE, --colortile COLORTILE
                        Color Tile size, 0 for no tile
  -st SRTILE, --srtile SRTILE
                        SR Tile size, 0 for no tile
  --tile_pad TILE_PAD   Tile padding
  -nosr, --no_superr    SR or not SR by RealESRGAN_x4plus_anime_6B
                        aftercolored
  -ca, --color_all      colorall images, no skip color one
  -onlysr, --only_sr    only SR all images, no color
  -sub, --all_subdir    handle all input sub folders

Samples

Original Colorization

🤗 Acknowledgement

Based on https://github.com/qweasdd/manga-colorization-v2

Thx https://github.com/xinntao/Real-ESRGAN

About

An amazing manga comic colorization project | 漫画 AI 上色

License:GNU General Public License v3.0


Languages

Language:Python 74.1%Language:Jupyter Notebook 25.9%