Awesome

Boruvka Superpixel

This code is a reimplementation of the superpixel segmentation based on Boruvka's minimum spanning tree algorithm: Xing Wei, Qingxiong Yang, Yihong Gong, Narendra Ahuja, and Ming-Hsuan Yang: Superpixel Hierarchy, IEEE Transactions on Image Processing 27, 4838 (2018) journal arXiv

build python module

• make module (in root directory of the repo)
  this creates pybuild/boruvka_superpixel.*.so, which is to be imported from python

test python module

• ./src/boruvkasupix.py <input_img_path> <output_img_path> <n_supix>
  example:
  • ./src/boruvkasupix.py test.png test_out.png 100

library interface

• c++ and python
• Data types supported: uint8, uint16, int8, int16, int32, float32, float64.
  In all cases the internal data type is float32 as a compromise between precision and memory use. The data type of the arrays feature and border_strength in build_2d() or build_3d() should be the same, but it can be independent of the data type of data in average().

dependencies

• python3
• numpy
• cython 0.28
• gcc/g++