Caffe is a deep learning framework made with expression, speed, and modularity in mind. It is developed by the Berkeley Vision and Learning Center (BVLC) and community contributors.
This branch is for Windows and Visual Studio 2013.
Most of procedures I did and 3rdparty are from Neil Shao's blog post. If you have any problems, you may refer to the blog post. Most of issues including LMDB(+ memory-hog issue), pycaffe, matcaffe have been solved in this repository.
We need following external libraries to build caffe.
- CUDA 7.5 + cuDNN v3
- OpenCV 3.0.0
- Boost 1.56
- OpenBLAS (in 3rdparty)
- GFlags, GLog, Protobuf (in 3rdparty)
- LevelDB, LMDB, HDF5 (in 3rdparty)
You can download pre-built 3rd party libraries here. You should extract 3rdparty folder to caffe/.
https://www.dropbox.com/s/v038gymle57kn7d/3rdparty.zip?dl=0 (Modified in Sept. 16, 2015)
Check if following environment variables are set.
- CUDA_PATH_V7_5
- OPENCV_HOME (like
D:\dev\opencv) - BOOST_1_56_0 (like
D:\dev\boost_1_56_0) - Path (add like
D:\dev\opencv\build\x64\vc12\bin;D:\dev\caffe\3rdparty\bin;C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\bin)
For pycaffe,
- PYTHON_27_HOME (like
C:\Python27) - PYTHONPATH (add like
D:\dev\caffe\python) - Path (add like
D:\dev\boost_1_56_0\lib64-msvc-12.0)
And, execute scripts/GeneratePB_python.bat. You should also install all python packages in caffe/python/requirements.txt. Most of packages can be installed by pip, but you may have to install some packages yourself. For leveldb, you may refer to https://code.google.com/p/py-leveldb/issues/detail?id=12.
For matcaffe,
- MATLAB_HOME (like
C:\Program Files\MATLAB\R2014a)
And, execute addpath matlab in MATLAB command line.
Open caffe/caffe.sln and set Release mode.
You may need some changes in project settings according to your environment.
Finally, you can build caffe.