rev-unet's Introduction

Reversible U-net for Medical Image Segmentation

This repos is revised based on https://github.com/RobinBruegger/PartiallyReversibleUnet
The course project require us to read on a paper and then revise and modify based on original project
Due to the time and compuational resource limit, I deprecated the original codes with following changes:
- Transfer from brain tumor segmentation to hippocampus segmentation
- revise pure dice loss to BCE loss and comibination of dice loss and BCE loss to compare impact of different loss
- add dialted convolution
- I also try to improve the result with some machine learning tricks like top-k loss
- I revise the original baseline model for more fair comparision(similar number of network parameters)
This repo is a implemented-from-scratch version and will move to a mutli-task topic in future
Results from report
- Numerical Results
- Visual Results(in slices)

Code Structure for Deep Learning

Data
- process_hdf5 save as hdf5
- process_json(tbd)
  - json output with images path and label path
Models
- utils - necessary function
  - maybe move evaluation metric here?
- network
  - no-new-net with different elemental blocks
- loss
- backbone network
  - network blocks
dataProcessing
- dataloader for train and test
Utils
- logger
- evaluation/metric
visualization - jupyter notebook
Trainer - APIs
- save/load weights
- lr scheduler
- optimizer
train
test
evaluation - evaluate predicted result
config - configurate parpameters

implement revtorch blocks by myself to try to improve
move to a
Ref and cite: @article{PartiallyRevUnet2019Bruegger,
author={Br{"u}gger, Robin and Baumgartner, Christian F. and Konukoglu, Ender},
title={A Partially Reversible U-Net for Memory-Efficient Volumetric Image Segmentation},
journal={arXiv:1906.06148},
year={2019},}

Recommend Projects