Caffe branch for R-FCN

This is a branch of Caffe supporting R-FCN, which has been tested under Windows (Windows 7, 8, Server 2012 R2) and Linux (Ubuntu 14.04).

Linux Setup

Pre-Build Steps

Copy Makefile.config.example to Makefile.config

We need to modify Makefile.config to specify some software PATHS, you may view my Makefile.config for reference.

CUDA

Download CUDA Toolkit 7.5 from nVidia website.

Matlab

Uncomment MATLAB_DIR and set MATLAB_DIR accordingly to build Caffe Matlab wrapper. Matlab 2014a and later versions are supported.

cuDNN (optional)

For cuDNN acceleration using NVIDIA’s proprietary cuDNN software, uncomment the USE_CUDNN := 1 switch in Makefile.config. cuDNN is sometimes but not always faster than Caffe’s GPU acceleration.

Download cuDNN v3 or cuDNN v4 from nVidia website. And unpack downloaded zip to $CUDA_PATH (It typically would be /usr/local/cuda/include and /usr/local/cuda/lib64).

Build

Simply type

make -j8 && make matcaffe

Windows Setup

Requirements: Visual Studio 2013

Pre-Build Steps

Copy .\windows\CommonSettings.props.example to .\windows\CommonSettings.props

3rd party dependencies required by Caffe are automatically resolved via NuGet.

CUDA

Download CUDA Toolkit 7.5 from nVidia website.

Matlab

Set MatlabSupport to true and MatlabDir to the root of your Matlab installation in .\windows\CommonSettings.props to build Caffe Matlab wrapper. Matlab 2014a and later versions are supported.

cuDNN (optional)

Download cuDNN v3 or cuDNN v4 from nVidia website. Unpack downloaded zip to %CUDA_PATH% (environment variable set by CUDA installer). Alternatively, you can unpack zip to any location and set CuDnnPath to point to this location in .\windows\CommonSettings.props. CuDnnPath defined in .\windows\CommonSettings.props. By default, cuDNN is not enabled. You can enable cuDNN by setting UseCuDNN to true in the property file.

Build

Now, you should be able to build .\windows\Caffe.sln

Remark

After you have built solution with Matlab support, copy all files in .\Build\x64\Release to R-FCN\external\caffe\matlab\caffe_rfcn.

Further Details

Refer to the BVLC/caffe master branch README for all other details such as license, citation, and so on.