Darknet on OpenCL

Darknet

Darknet is an open source neural network framework written in C and CUDA.
It is fast, easy to install, and supports CPU and GPU computation.
You can find the origin project here.

Darknet on OpenCL

As the origin darknet project is written in CUDA, this project is to port
the darknet to OpenCL. Also, darknet is assumed to run on Linux and used
some POSIX C APIs, This project rewrite this with standard C++ libraries.
So that it could also run on Windows.

Dependency

OpenCL
Make sure you have OpenCL installed, and set environment variables OPENCL_SDK
point to your OpenCL installed path.
clBLAS
clBLAS is equivalent to cuBLAS, you can find the source code here
and compile it yourself.
or you can use binary library for Windows/Ubuntu x64 platform I have already provided
You can find clBLAS.lib/clBLAS.dll for Windows and libclBLAS.so for Linux
as well as header file here.

Build

Windows
This project is prepared with Visual Studio 2017, just open darknet_cl.sln
and build it.
To enable OpenCL, please set environment variables OPENCL_SDK first.
To enable OpenCV, please set environment variables OPENCV_INCLUDE_DIR
and OPENCV_LIB first.
Linux
mkdir build && cd build
cmake ../
make

Usage

Once you compiled this project, it has the same usage as darknet,
you can find it here.
If you compile the project depend on the clBLAS library I provided, you'd better
copy dependent library clBLAS.dll or libclBLAS.so to
system lib path.(C:\Windows\System32 or /usr/lib).

Attention

This project didn't build the DarkGo into the darknet_cl, maybe support it later.