Skip to content

PytorchConnectomics/uct_seg

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

uct_seg

MicroCT 3D volume segmentation toolset for PyTorchConnectomics workflows.

Demo

Segmentation outputs on a microCT volume:

Input All targets
Input All
Blood vessel Dendrite Soma
BV Dendrite Soma

Scope

Per-target sub-packages:

Module Purpose
uct_seg.bv Blood vessel segmentation
uct_seg.artifact Crack / bright-aggregate masking
uct_seg.nuclei Nuclei (cb) segmentation + curation
uct_seg.dendrite Large dendrite segmentation
uct_seg.celltype Cell-type classification (planned)

Shared infrastructure:

Module Purpose
uct_seg.util.io h5/png/txt I/O, VAST png ↔ uint32 seg, mkdir, dtype packing
uct_seg.util.vast VAST meta parsing, anchor-tree writer, good/bad relabel
uct_seg.util.seg_ops removeSeg, removeSmall, relabel, seg_iou3d, get_bb_label*
uct_seg.util.imops imAdjust, CLAHE
uct_seg.util.viz Neuroglancer layer helpers + viewer launcher
uct_seg.util.butterfly write_bfly_v2 JSON spec writer
uct_seg.slurm Parameterized SLURM script generator

scripts/ holds one-shot dataset prep pipelines ported from the lab's legacy ct/ codebase (NucMM/Nag, Dyer-17, Xiaotang synchrotron, neuroglancer upload).

configs/ holds PyTC YAML configs per target (thin wrappers — training runs through pytc CLI).

tests/ holds runnable smoke tests on synthetic data (no cluster paths required).

demo/ holds visual showcase images of segmentation results — see demo/README.md.

docs/data_paths.md lists known cluster paths for source datasets.

Install

pip install -e .
# extras:
pip install -e ".[viz,cv,cc,pytc]"

Quick start

from uct_seg.util.io import readh5, writeh5, vast2Seg, seg2Vast
from uct_seg.util.vast import vastMetaRelabel, writeVastAnchorTreeById
from uct_seg.util.seg_ops import seg_iou3d, removeSeg
from uct_seg.util.viz import ngLayer3, launch_viewer

Run a dataset prep:

python scripts/prep_nucmm.py 8       # quantize + downsample images
python scripts/prep_nucmm.py 9       # create neuroglancer precomputed
python scripts/prep_xt_ct.py 0.1 0 4 # quantize tiffs, 4-way parallel

Generate SLURM jobs:

uct-slurm-gen nag --num 10 --cmd "python scripts/prep_nucmm.py 0"

Status

Target Status
io / vast / seg_ops / imops / viz / butterfly / slurm DONE (ported)
nuclei.postproc / nuclei.curate DONE (ported)
artifact.mask DONE (ported)
bv CURRENT (preproc stub, training via pytc config)
dendrite TODO
celltype TODO

About

microCT segmentation

Resources

License

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages