Adding another predicting class branch

[mamunir]

I have successfully trained on custom dataset with Mask RCNN. We have two modules for _cls and _bbox prediction, while training there is no need to disturb already set Detectron module. Now I want to add another classification module along with _cls and _bbox for jointly training, for that where should i make changes in the whole pipeline and where to define new classes. Moreover, where to set parameter (scaling factor for all losses) when back-propagate in a combined fashion?

Hope to hear soon! Best

[mamunir]

@ir413 , in addition to previous, for caffe, we can use prototxt to define. please guide in detectron