Implementing foundation models

Is your feature request related to a problem? Please describe.
I have been using MONAI for my project the last few months, the main problem is I need to use publicly available foundation models such as RAFT or MiDaS to generate intermidiate inputs for my models. This makes the implementation of such networks a bit tricky as you have to stick to certain input requirements (size, normalization, etc.) and implement those outside the MONAI bundles.

Describe the solution you'd like
Implementing a set of foundation models for MONAI. For my own application, I need depth estimation and optical flow estimation. However, I think it will benefitial to add models for segmentation and object detection also. I think there are two possible solutions:

1. Create native foundation models in MONAI for these tasks.
2. "Porting" some of the general foundation models available in Pytorch such as RAFT or MiDaS and implement them inside the MONAI framework.

Additional context
I'd like to use this post as a discussion about the implementation of such foundation models. I am more than happy to participate on the process, as it would massively benefit my own work.

The issue names no files, tests, or entry points and presents foundation-model support as an open discussion. First narrow the scope to specific models and tasks, then define the integration point and acceptance criteria before implementation can be considered complete.