setanarut / decompose

Decompose image into layers (Fast Soft Color Segmentation)

Decompose

Fast Soft Color Segmentation without Residue predictor. Paper PDF

Installation

pip install git+https://github.com/setanarut/decompose 

Example

from decompose.decomposer import decompose
from PIL import Image

img = Image.open("image.jpg")
layers = decompose(img)

for layer in layers:
    print(layer)

# Decomposer mask generation...
# Decomposer processing alpha layers...
# Decomposer Done!
# <PIL.Image.Image image mode=RGBA size=904x368 at 0x1144F71D0>
# <PIL.Image.Image image mode=RGBA size=904x368 at 0x111EA4A90>
# <PIL.Image.Image image mode=RGBA size=904x368 at 0x114585510>
# <PIL.Image.Image image mode=RGBA size=904x368 at 0x114587790>
# <PIL.Image.Image image mode=RGBA size=904x368 at 0x1145856D0>
# <PIL.Image.Image image mode=RGBA size=904x368 at 0x1145858D0>
# <PIL.Image.Image image mode=RGBA size=904x368 at 0x114585C90>

You can write layers as ORA file

from decompose.utils import images_to_ORA

images_to_ORA(layers).save("img.ora")

If palette is not given as an argument, a palette is created automatically. Colors can be selected manually.

manual_palette = [
    (255, 255, 255),
    (3, 135, 3),
    (3, 193, 160),
    (1, 167, 255),
    (255, 243, 0),
    (193, 0, 0),
    (3, 0, 2),
]
layers = decompose(img, manual_palette)

decomp command-line tool example

Saves layers as ORA (Open Raster) file. It can be opened with Krita. Also saves the palette.

$ decomp ~/Desktop/img.png

# Decomposer mask generation...
# Decomposer processing alpha layers...
# Decomposer Done!
# ORA saved: img.ora
# 7
# Palette saved: img_palette.png

Note

The model only supports 7 colors. The same color can be repeated for less color.