Classification of the HAM-10000 dataset on skin lesions

Challenge 1 The binary problem of classifying Nevus images vs all the others. We will give you 6000 images, 3000 being nevus, 3000 being a combination of the others to train the system. The test set will consist on 1015 images.

Challenge 2 A three-class problem consisting on the classification of melanoma vs benign keratosis vs basal cell carcinoma. The training set will consist on 1000 images for the first two classes and 500 for the last one (imbalanced problem). The test set consist on 226 images

Hair removal with black-hat transformation

1. Hu moments
2. LBP
3. Color Histogram
4. HoG
5. GLCM
6. Haralick
7. SIFT

1. Data Normalization
2. PCA

1. ResNet
2. DenseNet
3. VGG
4. AlexNet