How can you use a Transformer model as a backbone? No documentation available

Hey,
Just wanted to know that how could you use a pre-trained Transformer model for object detection? I want to use [Microsoft's DiT](https://github.com/microsoft/unilm/tree/master/dit/object_detection). Seems like they have used it with `Detectron2` but no clear steps have been described. Can you please let me know about where can I find the blog / tutorial / documentation on how to fine tune that transformer architecture to use it as a detection backbone with `Faster RCNN` / `Cascade RCNN`.

Thanks in advance :)

Start with Microsoft's DiT object_detection example and the existing Detectron2 documentation. Determine and document the steps for fine-tuning the Transformer architecture as a detection backbone with Faster R-CNN and Cascade R-CNN; done means a newcomer can follow the guide without needing undocumented setup knowledge.

Written by the indexing model from the issue text.