Caffe: Convolutional Architecture for Fast Feature Embedding

摘要: Caffe provides multimedia scientists and practitioners with a clean modifiable framework for state-of-the-art deep learning algorithms collection of reference models. The is BSD-licensed C++ library Python MATLAB bindings training deploying general-purpose convolutional neural networks other models efficiently on commodity architectures. fits industry internet-scale media needs by CUDA GPU computation, processing over 40 million images day single K40 or Titan (approx 2 ms per image). By separating model representation from actual implementation, allows experimentation seamless switching among platforms ease development deployment prototyping machines to cloud environments.Caffe maintained developed the Berkeley Vision Learning Center (BVLC) help an active community contributors GitHub. It powers ongoing research projects, large-scale industrial applications, startup prototypes in vision, speech, multimedia.