Swap weights between mask-rcnn-101 and PointRend

[pixml27]

Hi all!
I have a mask rcnn model (mask_rcnn_R_101_FPN_3x) trained on 10 classes. The model works well, f-score is about 90%. I used 7,000 images to train it, but the annotations wasn't perfect (the edges of the objects were not neatly marked).
Now I need my model to predict object edges very accurately and I want to use a PointRend.
But for PointRend training, I have about 700 images with very accurate annotations. If I will train it on only 700 images, then model F-score will definitely drop a lot.
So, what I want to do is to train the point-rend on 700 images, but at the start not take the weights trained on coco - I want load into the backbone and other heads (except PointRendMaskHead) the weights from the mask rcnn trained on 7000 images.
I can do this by writing the name of each layer manually, but i have to write a lot:
pointrend_model.backbone.fpn_lateral2.weight = maskrcnn_model.backbone.fpn_lateral2.weight
Do you know how to do it automatically (iterate over the names of the model layers)?
Moreover, do you think this approach will work?