How to understand the "training for multiple flow" and the "div_flow"

[poppinjie]

div_flow is used for 2 reasons :

Training is deemed more stable for low output target values.
it was used by initial caffe code, and thus it makes this training compatible with a pretrained model from caffe.

I try to answer my own question. HaHa! I referred to this answer.
How to understand the "training for multiple flow" and the "div_flow" #28

[Clément Pinard]

Hi, what is your question exactly ?