Computer Vision News - February 2017

The unique properties of the proposed network are: a) efficient hybrid training scheme, utilizing dense training as proposed by Long et al. in 2015; b) implementing the insight developed in 2D networks, that deeper networks with smaller kernels are more discriminative in 3D, to provide improved performance; c) a dual-pathway network, which incorporates both local and larger contextual information; d) post-processing of the network’s soft segmentation, with CRF for improved results. Trying it out: To use DeepMedic, you need a Linux machine with a GPU card. If you don't have the needed hardware, you can use the AWS service: a machine with the required specifications costs about $1 per hour. We tried DeepMedic on AWS, with a p2.xlarge machine with Ubuntu-16.04 and disk size of 20GB. The following libraries were installed: Using the following commands: Once installation is complete, use the following three commands to verify that everything works as expected: Computer Vision News We Tried for You 9 Tried for You  Theano : Deep Learning library  Nose : Theano’s unit tests  NiBabel Library for loading NIFTI files  Parallel Python : Library used to parallelize process  Scipy : For statistics, optimization, integration, algebra, machine learning  Numpy : General purpose array-processing package 1. sudo apt-get install python python-setuptools python-pip python-dev build-essential 2. sudo pip install --upgrade pip 3. sudo apt-get install python-numpy libblas-dev liblapack-dev libatlas-base-dev gfortran 4. wget https://developer.nvidia.com/compute/cuda/8.0/prod/local_installers/cuda-repo- ubuntu1604-8-0-local_8.0.44-1_amd64-deb 5. sudo dpkg -i cuda-repo-ubuntu1604-8-0-local_8.0.44-1_amd64-deb 6. sudo apt-get update 7. sudo apt-get install cuda 8. export CUDA_HOME=/usr/local/cuda-8.0 export LD_LIBRARY_PATH=${CUDA_HOME}/lib64 PATH=${CUDA_HOME}/bin:${PATH} export PATH ./deepMedicRun -newModel modelConfig.cfg ./deepMedicRun -train trainConfig.cfg -model tinyCnn.save ./deepMedicRun -test testConfig.cfg -model tinyCnn.save