How to train ROB SceneFlow?

Hi,I have not found the prototxt about training the ROB SceneFlow,Can you provide the corresponding prototxt?
In this project,4 dataset was used in the  train_ROB_stage_one.prototxt,and 3 dataset was used in the 
train_ROB_stage_two.prototxt,or maybe the ROB SceneFlow model is the output of stage1?
