SKOOTS

Skeleton Oriented Object Segmentation

SKOOTS is an approach at high resolution, 3D mitochondria segmentation. Bridging the gap between boundary-prediction networks and flow based approaches, SKOOTS offers robust segmentation of high resolution mitochondria in previously difficult situations.

SKOOTS is presented as a python library with documentation provided here. Created with a functional interface, see the API Flow guide for instructions on how to incorporate a SKOOTS into a segmentation pipeline, and our example scripts for implementation.

Basics

• Quickstart
• Installation
• Train
• Inference

Tutorials

• Training a SKOOTS Model
  • Prepare Your Data
  • Precompute Ground Truth Skeletons
  • Configure your training
  • Train your model
• Inference with SKOOTS
  • Inference in the CLI
  • Convert Outputs
• Detailed Training Protocol
  • Imports
  • Setup DataDistributedParallel
  • Data Loading and Augmentation
  • Optimizers, Schedulers, Loss
  • Sigma
  • Vector Scaling
  • Calling the DataLoader and a Simple Training Iteration
  • Warmup
  • Main Training Loop
  • Save the model
• Detailed Inference Protocol
  • Imports
  • Load the image and model
  • Preallocate intermediary arrays
  • Iterative Evaluation
  • Assigning ID’s
  • Generate Instance Masks

API Flow

• Evalutation API Flow
• Training API Flow
• Data Transforms API Flow

API Reference

• skoots.lib
  • skoots.lib.embedding_to_prob
  • skoots.lib.vector_to_embedding
  • skoots.lib.flood_fill
  • skoots.lib.skeleton
  • skoots.lib.morphology
  • skoots.lib.utils
  • skoots.lib.mp_utils
  • skoots.lib.merge
• skoots.train
  • skoots.train.dataloader
  • skoots.train.distributed
  • skoots.train.engine
  • skoots.train.generate_skeletons
  • skoots.train.loss
  • skoots.train.merged_transform
  • skoots.train.setup
  • skoots.train.sigma
  • skoots.train.utils
  • skoots.train.erosion
• skoots.validate
  • skoots.validate.compare
  • skoots.validate.lib
  • skoots.validate.stats
  • skoots.validate.utils