Faster rcnn with resnet and sift

Hi everyone,

I have a pytorch implementation of faster rcnn with resnet50 and fpn. I was doing some research and learned about feature descriptors and i found out about SIFT and i believe it can be a great addition to my model.
At this point in time i m not sure how to implement it into my architecture to get the model to learn these additional features. I would really appreciate the help and guidance on this matter

Thanks in advance