This project merge Caffe-C3D from facebook C3D and Caffe-windows from initialneil. The readme below is from initialneil.

Caffe is a deep learning framework developed with cleanliness, readability, and speed in mind.
Consult the project website for all documentation.

This is not the official Caffe but a repository to compile Caffe in Windows with Visual Studio 2013 + CUDA 6.5 + OpenCV 2.4.9.

1. Download the repository.
2. Download 3rdparty dependencies from here, and unzip to the root of caffe. Make sure folder '3rdparty', 'bin', 'caffe', 'examples', 'include', ... are in the same path.
3. Install CUDA toolkit 6.5.
4. Install OpenCV 2.4.9 and set up the system variable OPENCV_X64_VS2013_2_4_9 to the root of OpenCV.
5. Install Boost 1.56 and set up the system variable BOOST_1_56_0 to the root of Boost.
6. Open caffe/caffe.sln in Visual Studio 2013 and BUILD. caffe.exe will be generated in caffe/bin.

1. run data/mnist/get_mnist.bat to get MNIST dataset.
2. run examples/mnist/create_mnist.bat to convert data to leveldb format.
3. run examples/mnist/train_lenet.bat to start training.

My step by step record can be found here.