* because it uses `numpy.flip` in `test_image.py`

* move net_op to operators

* init softmax grad op

* add compute code

* export Backward to python

* update test ,export op.type to python

* update python test, fix compute bug

* update unit test

* use eigen

* optimize eigen code

* add gpu test

* register softmax_grad GPU kernel and fix test bug

* typo

* follow comments

* Do not use `virtualenv` in Docker
* Split unittest into many target, make them parallel