IA Sante Challenge - Team Linkvalue

Challenge Home Page

Environment

This repo was made to run with :

• Python 3.5.2

• Keras 2.0.9

• Tensorflow 1.4.0

• Cuda 8.0

• Cudnn 5.1

Usage

You first have to edit the config file to set the basic parameters (location of your training data, future outputs...). And the, the generic usage of main.py file is as follow :

cd python/
python main.py <mode> <neural-net-name> (OPTIONAL : <weights-file.h5>) 

Train

cd python/
python main.py train <neural-net-name> (OPTIONAL : <weights-file.h5>) 

For instance, to run a train with inception V3 using weights of the model trained on ImageNet, just run python main.py train inception

Predict

To run a prediction, you need to pass a weights file as an argument.

cd python/
python main.py predict <neural-net-name> <weights-file.h5>

For instance, to run a predict with VGG using weights of the model you just trained, just run python main.py predict vgg /sharedfiles/outputs/model/vgg.h5

Preprocessing (ROI & data augmentation)

Before runing the trains, we ran some preprocessing our images using the preprocessing file like this :

cd python/
python preprocessing.py <input_folder> <output_folder> <im_size> <pool_size>

Some sources that inspired us

Find some documentation useful to start with here