kakumarabhishek / Illumination-based-Transformations-Skin-Lesion-Segmentation

This is the code corresponding to our CVPR ISIC 2020 paper.

Illumination-based Transformations Improve Skin Lesion Segmentation in Dermoscopic Images

This is the code corresponding to our CVPR ISIC 2020 paper. If you use our code, please cite our paper:

Kumar Abhishek, Ghassan Hamarneh, Mark S. Drew, "Illumination-based Transformations Improve Skin Lesion Segmentation in Dermoscopic Images", The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) ISIC Skin Image Analysis Workshop, 2020.

The corresponding bibtex entry is:

@InProceedings{Abhishek_2020_CVPRW,
author = {Abhishek, Kumar and Hamarneh, Ghassan and Drew, Mark S.},
title = {Illumination-based Transformations Improve Skin Lesion Segmentation in Dermoscopic Images},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) ISIC Skin Image Analysis Workshop},
month = {June},
year = {2020}
}

Dependencies

• PyTorch
• NumPy
• PIL
• SciPy (==1.1.0; for imread and imresize)

We also need to convert between the RGB and HSV color spaces, and scikit-image provides this functionality. However, scikit-image has issues with NumPy^1,2,3, and therefore to avoid installing it, we have borrowed the necessary functions from scikit-image source code and put them in skimg_local.py.

Usage

An example usage is shown in Transformations.ipynb, and three sample skin lesion images from the ISIC Archive are included in the images directory.