Is the pre-trained Resnet-50 object detection model available?

[VSJMilewski]

Hi, thank you for the nice repository! I had a question regarding the training of the VisualGenome models. The following is written in the paper:

"The object detection model is a ResNet50-FPN detector trained on Visucal Genome"

I have two questions:

1. I cannot figure out where I can find the pretrained weights of this model? Or are we expected to train this ourselves?
2. Why is the Backbone resnet different for VisualGenome and OpenImages?

Thanks for the help!