Skip to content

Latest commit

 

History

History
87 lines (67 loc) · 3.2 KB

README.md

File metadata and controls

87 lines (67 loc) · 3.2 KB

ROBUST MELANOMA-SCREENING

This repository contains the implementation of three techniques for automated melanoma screening (a Baseline, BossaNova and Deep Learning + Transfer Learning). We have compared them in the paper Towards Automated Melanoma Screening: Proper Computer Vision & Reliable Results.

You can find more details about the concepts and parametrization of each technique at the paper. The codes available in this repository are already parametrized according to the details provided in the text.

TABLE OF CONTENTS

  • Dependencies
  • Installation
  • Repository contents
  • Running

DEPENDENCIES

Note: we assume that the codes available here will be run on an Unix OS. You will need:

  • ImageMagick;
  • Matlab;
  • VLFeat for Matlab;
  • Java;
  • Python;
  • LIBLINEAR: already provided in this repository;
  • LIBSVM: already provided in this repository.

INSTALLATION

Please, follow the download and installation guidelines of each dependency.

For LIBSVM, it will be necessary to generate the executables through the command "make". Please, follow the guidelines at each README file and ignore the warning messages (if any).

After generating the LIBSVM executables, browse to the directory [setup] and allow the execution of the script "1_enableExecutables.sh":

chmod +x 1_enableExecutables.sh

Then, run the script. It will give execute permissions to all programs/scripts in this repository.

RUNNING

  1. Preprocessing: run script "../resources/convertEDRAAtlas.sh" to copy the images from EDRA CD-ROM to the "../dataset/images" folder. It also removes the images' black frames (if any);

  2. Baseline: follow the usage indications provided by running "baseline.sh" without arguments;

  3. BossaNova: follow the usage indications provided by running "bossanova.sh" without arguments;

  4. Deep+Transfer: follow the instructions provided at the specific README file.

REPOSITORY CONTENTS

baseline: 					
	code: 					baseline source code. 
	baseline.sh: 			main script to run experiments with baseline code. 
bossanova: 					
	code: 					BossaNova source code. 
	bossanova.sh: 			main script to run experiments with BossaNova code. 
dataset:					directory to store the images of "Baseline" and "BossaNova" experiments. 
	folds: 					directory containing the fold files of each validation protocol. 
deep-transfer: 				
	datasets:				the folder to organize datasets for "deep + transfer" experiments.
	resources:				stores the "deep + transfer" code dependencies. 
	source:					"deep + transfer" source code. 
	download_model.sh:		a script to download the VGG-M model file. 
	READ_ME.txt:			an explanation about how to run the deep learning code
resources:					
	libsvm-2.9_titans: 		a modified version of LIBSVM. See its README file for details. 
	convertEDRAAtlas.sh:	script to preprocess (organize, convert and crop) the images from EDRA CD-ROM for the experiments of this paper.
setup: 						
	1_enableExecutables.sh: script to setup the running environment. 
README.md:	 				this file;