parenx

[anastassiavybornova]

parenx exploration. implementation was straightforward (it's basically just one pip install and then 1 line of bash, one each for skeletonization and voronoi). the output files are quite large (doubling to quadrupling in size v the input files) so not commiting them for now. one more time realizing how cool it is to have our manual "ground truth", next step i'd like to implement is to automate plotting of all the cases from the protocol, comparing manual and parenx simplification. (then we can start thinking about how to quantify the comparison)

[martinfleis]

Does it have any parameters?

the output files are quite large (doubling to quadrupling in size v the input files)

That is expected give the result is an outcome of tessellation. You need to segmentize the geometry to get a reasonably close result to the original geoms. But it is a good point and could be seen as a limitation of the method.

[anastassiavybornova]

re parameters: yes, there are the following parameters (default values):

• skeletonization: tolerance (0m), line buffer (8m), raster scale (1)
• voronoi: tolerance (0m), line buffer (8m), voronoi scale (5) it's not really documented, i just found the get_args functions here: https://github.com/anisotropi4/parenx/blob/main/src/parenx/skeletonize.py#L37 https://github.com/anisotropi4/parenx/blob/main/src/parenx/voronoi.py#L28 i've run all with default parameters for now

re output files: to be more specific, they contain 3 layers: the original input (so it's duplicating the data), the simplified network, and the "primal network" (simplified network nodes plus edges that do NOT conserve morphology). you're saying that the simplified network would need to be segmentized? not sure what that means/how to do it

[martinfleis]

you're saying that the simplified network would need to be segmentized? not sure what that means/how to do it

No, I am saying that the Voronoi-based algorithm in parenx has done that.