Currently the binary library is only precompiled for for 32-bit and 64-bit OSX (thanks to Gene Kogan). Additional contributions for face recognition come from Antimodular Research.
After cloning the repository, you will need to run ./setup.sh to download the pre-trained ImageNet classifier.
After installing CUDA and recompiling ccv with CUDA support, replace ccv.a in the libs folder with your new static library. Then make a copy of the following CUDA libraries adjacent to ccv.a: libcublas_static.a, libcudart_static.a, libculibos.a. Rename them all to remove the lib prefix. Include them in the linking step. Finally, add GSL to the linker step, too. On OSX after installing GSL with Homebrew, this means adding -L/usr/local/Cellar/gsl/1.16/lib and -lgsl. After all of this, classification of a single 640x480 image should drop from 750ms to 400ms.