Builing object detection fine tunining and testing model (Faster R CNN FPN) in detectron2

Hello, As a small part of my project I want build faster R-CNN FPN network in detectron 2 for fine tuning and testing the object detection ctyscape dataset. I have pth file for the resnet 50 encoder, which I want to put into faster R CNN model of detectron and froze the model and only tune the other parts(FPN) of faster R-CNN to know how good my representation is of encoder Resnet 50. Then I want to take same model for testing. My dataset is cityscape.