santiagosiordia5 / esfpnet

Official Implementation of "ESFPNet: efficient deep learning architecture for real-time lesion segmentation in autofluorescence bronchoscopic video"

ESFPNet

Official Implementation of "ESFPNet: efficient deep learning architecture for real-time lesion segmentation in autofluorescence bronchoscopic video"

Global Rank

Architecture of ESFPNet

Installation & Usage

Enviroment (Python 3.8)

• Install Pytorch:

conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch

• Install image reading and writting library:

conda install -c conda-forge imageio

• Install image processing library:

pip install scikit-image

• Install library for parsing and emitting YAML:

pip install pyyaml

• Install other packages:

conda install pillow numpy matplotlib

• Install Jupyter-Notebook to run .ipynb file

conda install -c anaconda jupyter

Dataset

• Download the training and testing dataset from this link: Experiment Dataset
• Extract the folders and copy them under "Endoscope-WL" folder
• The training and testing dataset are ordered as follows in "Endoscope-WL" folder:

|-- TrainDataset
|   |-- CVC-ClinicDB
|   |   |-- images
|   |   |-- masks
|   |-- Kvasir
|       |-- images
|       |-- masks

|-- TestDataset
|   |-- CVC-300
|   |   |-- images
|   |   |-- masks
|   |-- CVC-ClinicDB
|   |   |-- images
|   |   |-- masks
|   |-- CVC-ColonDB
|   |   |-- images
|   |   |-- masks
|   |-- ETIS-LaribPolypDB
|   |   |-- images
|   |   |-- masks
|   |-- Kvasir
|       |-- images
|       |-- masks

• The default dataset paths can be changed in "Configure.yaml"
• To randomly split the CVC-ClincDB or Kvasir dataset, set "if_renew = True" in "ESFPNet_Endoscope_Learning_Ability.ipynb"
• To repeat generate the splitting dataset, previous generated folder shold be detelted first
• To reuse the splitting dataset without generating a new dataset, set "if_renew = False"

Pretrained model

• Download the pretrained Mixtransformer from this link: Pretrained Model
• Put the pretrained models under "Pretrained" folder

Citation

If you think this paper helps, please cite:

@article{chang2022esfpnet,
  title={ESFPNet: efficient deep learning architecture for real-time lesion segmentation in autofluorescence bronchoscopic video},
  author={Chang, Qi and Ahmad, Danish and Toth, Jennifer and Bascom, Rebecca and Higgins, William E},
  journal={arXiv preprint arXiv:2207.07759},
  year={2022}
}

Since the training of MixTransformer based network requires a good GPU. One helpful state-of-the-art work compared in this paper without using MixTransformer backbone is CARANet If you also think this work helps, please cite:

@inproceedings{lou2021caranet,
author = {Ange Lou and Shuyue Guan and Hanseok Ko and Murray H. Loew},
title = {{CaraNet: context axial reverse attention network for segmentation of small medical objects}},
volume = {12032},
booktitle = {Medical Imaging 2022: Image Processing},
organization = {International Society for Optics and Photonics},
publisher = {SPIE},
pages = {81 -- 92},
year = {2022},
doi = {10.1117/12.2611802}}