A repository for building a GPU accelerated DeepCNN framework

##Prerequisites

• MatIO (a library for writing and reading Matlab .mat files)
• Cuda 7.5
• CuDNN v3
• CMake
• G++
• GTest (for system tests)
• CImg (for reading images)
• Git
• Octave

##Build

• Clone the repository from github:

git clone https://github.com/slundqui/DeepCNN.git

• Run CMake

ccmake .

• Fill out appropriate fields
• Press g
• Press c
• Compile

make

##Run Download CIFAR-10 database. Run script for formatting mat file, editing file to correct path

octave <repoBaseDir>/scripts/formatCifar.m

Executables in /bin:

• test: The testing suite for the toolkit
• cifarTrain: Executable for training
• cifarTest: Executable for testing, loading weights from cifarTrain
• cifarTest_IO: Executable for testing using weights from epoch 15. Writes out each estimate and ground truth for confusion matrix generation.