For each skin lesion image, I have applied Gaussian filter with Ostu threshold and then morphological filter (dilation_ with contour using OpenCV to automatically segment lesion object such as:
The task is to calculate the Dice Similarity (DS) Score of skin lesion dataset where M is the segmented lesion mask from Task 1 and M is the ground truth provided in the dataset here.
The equation to calculate DS score is:
To get the object segmentation, run automated_legion_detector.py and for DS score, run the dice_similarity_score.py with the dataset. The mean DS score is 0.8 with std. dev of 0.812 which is relatively better considering the data was not completely clean.