README.md

Superpixel Clustering for Image Segmentation

Code started with Superpixel Hierarchical Clustering algorithm (SPHC) For Image Segmentation repository, but changed a lot, using SLINK algorithm, and became much more faster.

Superpixel Segmentation

Superpixel segmentation methods

Hierarchical Methods for Superpixel Segmentation

Partition hierarchy of segmentation methods

Goals

• New hierarchical superpixel clustering method;
• Superpixels comparison;
• Results tested on BSDS500 dataset;
• Jupyter notebook codes, with images and examples;

Prerequisites

Conda Environment

You can use Conda to configure your environment. Conda file with all prerequisites are available here

conda env create -f i2dl.yml

Code

Detailed Notebooks

• fmeasure-segb.ipynb - Creates hierarchical segmentation with superpixels algorithms;
• fmeasure.ipynb - Creates regular segmentation with superpixels algorithms;
• segmentation.ipynb - Provides example of the developed method;
• test_groundtruth.ipynb - Notebook just to test groundthruth evaluation method;
• plot.ipynb - Plot results of hierarchical and regular segmentation methods;

Paper

Contains a paper (unpublished) showing the results of the superpixels algorithms and comparing them. Available only in portuguese.

References

Available in bib folder or inside the paper.