boun7yhunt3r / melanoma-detection Goto Github PK

This project forked from apolanco3225/melanoma-detection

Applying Mask R-CNN to the segmentation of skin lesions such as Melanoma (skin cancer), seborrheic keratosis and nevus. The model receives an input image and then feed it through the Mask R-CNN network to obtain the mask prediction. The predicted mask is only 15 × 15 pixels so is applied nearest neighbor interpolation to resize it back to the original image dimensions. The resized mask can then be overlaid on the original input image and then be applied the input image of a model that classifies among the different classes of skin lesions, helping improve the accuracy. Finally different models are created to use ensemble methods in order to develop a super model to make the final classification decision using every individual model previously created.