A lightweight Python software for semi-supervised segmentation on images, it can be utilized for tasks such as image segmentation, background extraction, semantic segmentation, colorization, etc. The code, written in pure Python, can be easily integrated with Flask or Django backend for a web app.

Image w annotation	Segmentation

Segment 1	Segment 2

pip install pisegment

• Command line

pisegment --input "path/to/image/tobe/segmented" --mask "path/to/the/generated/annotation"

• The above command opens up a GUI for image annotation. Use right click drag for annotation. There are 9 different color options. Press 1 for black color, 2 for blue color and so on. After annotating, press s to let the segmentation begin. Zoom in/out with mouse wheel and left click drag for panning image.

• By default, the denoising filter is on. If your image doesn't have noise, turn it off using the --no_filter option. You may also use this option if you are already using different software for denoising.

pisegment --input "path/.." --mask "path/.." --no_filter

• For fast processing, consider downsizing your image to under 256 X 256 to generate the segmented mask, then upscale the segmented mask to the original size.
• For complex images, like the last example in the Results, to get a precise segmentation, the parameter --sig plays a crucial role. See the Jupyter_demo for the workflow on how to segment such kind of images.

Behind the scenes, it is basically using a Dijkstra like algorithm to propagate the labels on graphs. The code is very similar to the Algorithm 2 of the following paper: https://hal.science/hal-00932510/document

The following description of parameters is useful for achieving good segmentation.

pisegment --input "path/.." --mask "path/.." --sig 1.00e-02  --k 4 --k_ 10 --ps 3 --no_filter