Skip to content

Latest commit

 

History

History
53 lines (46 loc) · 2.35 KB

README.md

File metadata and controls

53 lines (46 loc) · 2.35 KB
Raw

Adapted NuClick for generating ground truth for the OCELOT 2023 Challenge

This repository provides the NuClick code (link to original repo) as adapted for generating segmentation ground truth maps for the OCELOT 2023 Challenge.

Idea

NuClick enables nuclei segmentation with the help of a guiding signal. In our case the guiding signal is the cell point annotation from the OCELOT train data. This way we obtain ground truth segmentation maps for the OCELOT 2023 Challenge by using NuClick (link to NuClick paper ).

Original image Cell point annotations
(OCELOT ground truth)		 NuClick prediction
alt text alt text alt text

Requirements

Keras==2.2.4 h5py opencv-python-headless==4.1.2.30 pandas==0.25.1 matplotlib==3.1.1 six tensorflow-gpu==2.4.0 scipy numpy albumentations==0.3.1 scikit-image Pillow==8.0.1

Inference:

Download weights for nucleus segmentation from here and save it inside weights folder:

  • Define directories in the config.py: path to annotation csvs: mat_path = '' , path to images: images_path = '' and save_path = ''
  • run test_all_images

Credits

@article{DBLP:journals/corr/abs-2005-14511,
  author       = {Navid Alemi Koohbanani and
                  Mostafa Jahanifar and
                  Neda Zamani Tajadin and
                  Nasir M. Rajpoot},
  title        = {NuClick: {A} Deep Learning Framework for Interactive Segmentation
                  of Microscopy Images},
  journal      = {CoRR},
  volume       = {abs/2005.14511},
  year         = {2020},
  url          = {https://arxiv.org/abs/2005.14511},
  eprinttype    = {arXiv},
  eprint       = {2005.14511},
  timestamp    = {Wed, 03 Jun 2020 11:36:54 +0200},
  biburl       = {https://dblp.org/rec/journals/corr/abs-2005-14511.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}