CodePudding user response:
Caffe2 can be regarded as a caffe finer-grained refactoring, on the basis of practical, increase the extensibility and flexibility ofCaffe of Net consisting of granularity coarser layer, each layer of the weight and bias in storage layer level, is very simple and intuitive, but for specific platforms do optimization, would be tedious, not only have to add the layer of concrete optimized implementations, and modify the caffe. Proto file and recompile
Caffe2.net component, adopting tensorflow, mxnet operator use of these frameworks, such as by more granular operator, when particle size smaller, can do more optimization.