ISIC CLassification

Create virtual Environment and Install dependencies

    /path/to/project/directory
    conda create -n isic
    conda activate isic
    pip3 install -r requirement.txt

Download Data

    /path/to/project/directory
    mkdir DATA

using this path https://www.kaggle.com/datasets/salviohexia/isic-2019-skin-lesion-images-for-classification/data Download the data, place it in DATA directory.

About the Code

1- To Avoid writing training loop, pytorch lightining was used. 2- Same CNN architecture was used for binary and multiclass classification. Only the losses were changed 3- In dataloader.py, binary and multiclass labels are handeled. 4- In unils.py, some necessary functions such as plots are placed. 5- To ease the proces of logging the experiments, TensorBoard is used in the code.

Run the Experiments

you can run original.ipynb you can use defaulf setting to run the experiments or you cand change them to your desired batch_size, number of epochs, ....