Melanoma_Classification

Repository to reproduce the results on SIM-ISIC Competition 2020 Dataset hosted on kaggle competitions

Steps performed:

Image Preprocessing:
- Hair Artifact Removal using Bottom Hat Filter and inpainting.
- Color Constancy corrections using Gray world and max RGB algorithms(Originally constructed by LincolnZjx)
Training CNN: Efficient Nets are trained in both ensemble and stand-alone manner
- Stratified Group K-Fold Cross validation
- Label Smoothing
- Loss: FOCAL LOSS
Testing CNN:
- Tested using both public and private test data hosted on kaggle competition

Metrics Used:

The same metric AUC which is given in kaggle is used to measure the performance of the model since the dataset is highly skewed.

PROGRESS

Name		Name	Last commit message	Last commit date
Latest commit History 19 Commits
.idea		.idea
Preprocessing_Utils		Preprocessing_Utils
__pycache__		__pycache__
.gitignore		.gitignore
2020_Challenge_duplicates.csv		2020_Challenge_duplicates.csv
CleanedTest.csv		CleanedTest.csv
CleanedTrain.csv		CleanedTrain.csv
Diag_Encodings.npy		Diag_Encodings.npy
EDA+Baseline.ipynb		EDA+Baseline.ipynb
Ext_2019.csv		Ext_2019.csv
Merged_1819.csv		Merged_1819.csv
Preprocess.py		Preprocess.py
README.md		README.md
anatom_site_general_challenge_encodings.npy		anatom_site_general_challenge_encodings.npy
ext_2018.csv		ext_2018.csv
main.py		main.py
sex_encodings.npy		sex_encodings.npy