Spatially Varying Color Distributions for Interactive Multi-Label Segmentation

An unofficial Python implementation of the paper Spatially Varying Color Distributions for Interactive Multi-Label Segmentation by Claudia Nieuwenhuis and Daniel Cremers. This implementation was done in collaboration with @unlikelymaths.

Get Started

• Install
  Please install this repository by:

    pip install ./SVCD/
  

• Data The required data consists of an RGB image and a suitable image with user-drawn scribbles. In the image with scribbles, black values and pixels with alpha<1 are assumed to be unknown, besides each individual color is treated as a class.

Usage

Please also have a look into samples.

from svcd.svcd import SVCDutils, SVCDSegmentation

image_path = ...
scribble_path = ...
save_path = ...

# load image and scribbles
image = SVCDutils.load_image(image_path)
scribble, colors = SVCDutils.load_segmentation_from_image(scribble_path)

# run algorithm
svcd = SVCDSegmentation(max_iter=1000)
segmentation, energies = svcd(image, scribble)  

# save segmentation
SVCDutils.save_segmentation_as_image(save_path, segmentation, colors)