Static map built only once and then reuse it constantly, easy to do optimization on the map, more efficient map
Every establish the use of dynamic map, it is not easy to optimize
Static map can be serialized to disk, you can save the entire network can be overloaded, very practical deployment
FIG previous dynamic code duplication is required
Dynamic map compared to a static view of code more concise
In tensorflow static figure conditionals and loops requires a specific syntax, pytorch only python syntax can be achieved
pytorch for research, tensorflow and caffe, caffe2 for deploying applications
1