search

CristianoPatricio / coherent-cbe-skin

Code for the paper "Coherent Concept-based Explanations in Medical Image and Its Application to Skin Lesion Diagnosis", CVPRW 2023.
14 stars 1 forks source link

Detailed Dataset Preprocessing and Configuration #1

Closed JetWings closed 4 months ago

JetWings commented 5 months ago

This is an impressive approach, and I appreciate the author's work. Could you please provide a detailed description of the dataset preprocessing and configuration?

CristianoPatricio commented 5 months ago

Hi @JetWings,

Thank you for your interest in our work! You can find all the .CSV files for the dataset splits and dermoscopic concepts here.

Our approach utilizes a hard attention mechanism based on segmentation masks. We use the DeepLabV3 model trained on HAM10000 to extract segmentation masks for the datasets (PH2 and Derm7pt) in our work, in addition to the ground-truth segmentation masks provided by PH2 and the annotated ones by ours for Derm7pt. The lesion segmentation mask (pre-processing stage of Figure 2) is obtained using i) the extracted binary segmentation mask by DLV3 or ii) the ground-truth segmentation mask provided by the datasets.

Feel free to email me (cristiano.patricio@ubi.pt), and I will send you a link to download all the preprocessed images along with the DLV3 masks.