Calculate Deep Convolution NeurAl Network on Cell Unit

We introduce CACU, a new deep learning algorithm framework for CNN which using binary method to reduce the consumptions in convolution calculating. CACU introduces bit data flow to fit the CPU platform. Using binary-weights and xnor methods to speed-up the convolution's computation on CPU device. GPU is also supported in CACU. CUDA version is implemented for accelerating large scale models' training and inference. CACU is a C++ library with no dependencies except Boost. CACU is not only developed for the CNN's usage in application, it's helpful for researchers to take an inner investigation of bit method in CNN. It's a fully open-source platform which is available on GitHub.