Skip to content

Abe404/localise_to_segment

Repository files navigation

Localise to segment: An investigation into the accuracy improvements provided by localisation steps for semantic segmentation.

Preprint now available: https://arxiv.org/abs/2304.04606v1

Install dependencies.

pip install -r requirements.txt

Installing textfig is more manual as no pip package is available.

bash install_texfig.sh

The plots (optional) also require latex to be installed.

sudo apt-get install texlive-xetex   

Download the datasets from the Medical Segmentation Decathlon:

bash download_datasets.sh

To prepare the other datasets (cropped and resized versions with 50 random samples).

bach data_prep.sh 

To train the networks (Will likely take a couple of days depending on hardware)

bash run_train.sh    

To compute results as csv and run t-test

bash results.py

To make plots

python make_plot_and_tables.py

About

Localise to segment: crop to improve organ at risk segmentation accuracy

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published