ViTDet changing img_size

## 📚 Documentation Issue

Hi, I'm curious if I could get more information on MAE/ViT input size and ViTDet `img_size`. 

I'd like to train a ViT 'b' backbone with MAE and use it to create a ViTDet model for size 448 images. Should the MAE pertaining also ideally be at size 448? And if I change `img_size` from 1024 to 448, what else needs to be changed?