Add ability add custom/modified backbone to FasterRCNN

[abhayraw1]

🚀 Feature

In the Faster RCNN bolt, it would be awesome to have the functionality to pass custom pl.LightningModules or torch.nn.Module backbone. In the current state, the constructor requires the backbone argument to be a str

https://github.com/PyTorchLightning/lightning-bolts/blob/c3b60de7dc30c5f7947256479d9be3a042b8c182/pl_bolts/models/detection/faster_rcnn/faster_rcnn_module.py#L33-L63

Motivation

I wanted to try some custom version of the ResNet model in the Faster RCNN pipeline and thought this functionality might help.

Alternatives

These are the two workarounds that I think would work:

1. Reassign the backbone manually. Similar to how we reassign the predictor head in torchvision tutorials - link.
2. Subclass the FasterRCNN and manually reassign the backbone

[github-actions]

Hi! thanks for your contribution!, great first issue!

[Borda]

@abhayraw1 that would be great, do you want to implement it?

[abhayraw1]

@Borda sure! Will drop a PR soon!

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.