Closed ProNoobLi closed 4 years ago
In the implementation. I saw that training return flow2-flow6, but why does need all of flow? It should be end2end training and only top layer of flow2 is enough.
Solved, it's multiscale training
In the implementation. I saw that training return flow2-flow6, but why does need all of flow? It should be end2end training and only top layer of flow2 is enough.