conda create -n zw python==3.8
pip install --editable .
representation: affinity -> segmentation -> region graph
-
end-to-end functions:
zwatershed(T_threshes=800, T_aff=[0.01,0.8,0.2], T_aff_relative=True, T_dust=600, T_merge=0)
-
step-by-step for parameter search
-
step 1: initial segmentation:
zw_initial(affs, affs_low, affs_high)
- steepest ascent:
zw_steepest_ascent(aff, affs_low, affs_high)
- divide plateaus:
zw_divide_plateaus(seg)
- find basin:
zw_find_basins(seg)
- steepest ascent:
-
step 2: merge region:
- get region graph:
zw_get_region_graph(aff, seg)
- merge (with or without mst)
zw_merge_segments_with_function(seg, rg_affs, id1, id2, counts, T_size, T_weight, T_dust, T_merge, T_mst)
- mean-affinity:
- get region graph:
- zwatershed:
do_zwatershed.py
- mean-affinity agglomeration:
mean_aff.py
- paper: http://arxiv.org/abs/1505.00249
- original C++ version: https://bitbucket.org/poozh/watershed
- cython version: https://github.com/jakirkham/zwatershed
- julia version: https://github.com/seung-lab/Watershed.jl
- ours: start from the cython version and modify according to the julia version (with a bug fix)