INSTALL

------------------
Dependencies
-----------------

- CUDA (>= 4.0) (Not required for the CPU-only feature extractor).

- HDF5 (>= 1.8.11)
  Download the source code http://www.hdfgroup.org/ftp/HDF5/prev-releases/hdf5-1.8.12/src/hdf5-1.8.12.tar.gz
  To install locally -
  $ ./configure --prefix=~/local/ --enable-threadsafe --with-pthread=/usr/include
  $ make
  $ make install

- Protocol Buffers (>= 2.5.0)
  Download the source code https://protobuf.googlecode.com/files/protobuf-2.5.0.tar.gz
  To install locally -
  $ ./configure --prefix=~/local/
  $ make
  $ make install

- libjpeg
  This is often present on standard operating systems since it is used by a lot of programs.
  It can be downloaded from http://libjpeg.sourceforge.net/

- This code uses C++11. Some features require gcc >= 4.6.

---------------------------
Hardware requirements
---------------------------
NVIDIA GPU(s).
Multi-GPU feature requires Fermi boards.

-------------------
Compilation
-------------------
Compiling the CPU-only feature extractor -
- Set the paths to the dependencies in `convnet/cpu/Makefile`.
- Run make in `convnet/cpu`.

Compiling the full package -
- Set the paths to the dependencies in `convnet/Makefile`.
- Set the compute capability for your GPU.
- Run make in `convnet`.

Add convnet/bin to your PATH variable (maybe in your .bashrc)

---------------------
Run test examples
---------------------
Examples of nets on MNIST and ImageNet are included in 
examples/mnist and examples/imagenet.
Follow the instructions in the respective README.md files.