Unexpected behaviour after creating a branch in a trained network.

[ravising-h]

I have trained Retinanet on my custom dataset and achieved good results. After which I thought to try a different multi task architecture. I have created a different semantic branch which joins retinanet at resnet res2 block. It has different names for rest of its branch. (Added ''seg_'' at beginning of every block). When I load retinanet weight it doesnt gives me the previous result it acts like it is not trained. While loading weight It gives me no error. I have not touched retinanet branch of the model just added a semantic backbone ( from res3) and head separately in the model.