Installation

Dependencies:

1. An Nvidia CUDA sm_20 capable graphics card and the CUDA SDK. SparseConvNet is tested using CUDA 7.5, but older versions might work too.
2. The OpenCV library
3. The Armadillo library
4. Google's Sparsehash library

To install dependencies 2-4 on Ubuntu 14.04:

sudo apt-get install libarmadillo-dev libopencv-core-dev libopencv-highgui-dev sparsehash

In Ubuntu 16.04, sparsehash has been replaced by libsparsehash-dev.

To test the CNN:

1. Put the CIFAR-10 .bin data files in the Data/CIFAR10/ folder
2. Execute "make cifar10 && ./cifar10"

Possible problems.

1. If you get the error "libdc1394 error: Failed to initialize libdc1394" from OpenCV, run sudo ln /dev/null /dev/raw1394