Skin Lesion Segmentation

Vinicius de Paulo Souza Ribeiro
Department of Computer Engineering and Industrial Automation
School of Electrical and Computer Engineering
University of Campinas

Contact me:
E-mail: vinicius.ribeiro1@gmail.com
LinkedIn: https://www.linkedin.com/in/vribeiro1/

Requirements:

• Docker
• Python 3.5

How to run?

• Create a docker container with image nvidia/cuda:9.1-devel-ubuntu16.04

nvidia-docker run --userns=host \
                  --shm-size 8G -ti \
                  -e OUTSIDE_USER=$USER \
                  -e OUTSIDE_UID=$UID \
                  -e OUTSIDE_GROUP=`/usr/bin/id -ng $USER` \ 
                  -e OUTSIDE_GID=`/usr/bin/id -g $USER` \
                  -v /path/to/repo/skin:/workspace/skin \
                  -v /path/to/data/isic2017:/datasets/isic2017 \
                  --name container_name \
                  nvidia/cuda:9.1-devel-ubuntu16.04 /bin/bash

• Update apt-get and install Ubuntu dependencies

apt-get update
apt-get install -y curl git wget python3 python3-pip python3-dev nano

• Install Python dependencies

cd workspace/skin
pip3 install -r requirements.txt

• Run the code

CUDA_VISIBLE_DEVICES=<gpu-id> python3 train.py with /path/to/config/file

Acknowledges:

The code available in the models package is based in the following repos:

• DeepLab V3+: https://github.com/jfzhang95/pytorch-deeplab-xception
• AutoDeepLab: https://github.com/MenghaoGuo/AutoDeeplab
• U-Net: https://github.com/learningtitans/isic2018-seg
• LinkNet: https://github.com/e-lab/pytorch-linknet
• RefineNet: https://github.com/thomasjpfan/pytorch_refinenet