This repo contains the code to train and evaluate FCN8 network as described in A Benchmark for Endoluminal Scene Segmentation of Colonoscopy Images. We investigate the use of Fully Convolutional Neural Networks for Endoluminal Scene Segmentation, and report state of the art results on EndoScene dataset.
You need to install :
- Theano. Preferably the last version
- Keras
- The dataset(http://www.cvc.uab.es/CVC-Colon/index.php/databases/cvc-endoscenestill/), needs user registration
- (Recommend) The new Theano GPU backend. Compilation will be much faster.
The architecture of the model is defined in fcn8.py. To train a model, you need to prepare the configuration in train file where all the parameters needed for creating and training your model are precised.
To train a model, use the command : THEANO_FLAGS='device=cuda0,floatX=float32' python train.py. All the logs of the experiments are stored in the result folder of the experiment.
David Vázquez, Jorge Bernal, F. Javier Sánchez, Gloria Fernández-Esparrach, Antonio M. López, Adriana Romero, Michal Drozdzal and Aaron Courville