monodepth2 VS packnet-sfm

I've used monodepth2 before and it's great: https://github.com/nianticlabs/monodepth2

How can a single camera do that? My experience with computer vision is fairly limited so I'm curious how that would work. My understanding is you need to be able to generate a point map, stereo vision, or some non-CV related method , e.g. radar like the pixel 4. 2D depth estimation can be done with a single camera in somewhat useful way but it's not a secure way (https://github.com/nianticlabs/monodepth2 -- now somewhat similar functionality in OpenCV). Can you expand on what AI the single camera is being combined with that provides security guarantees?

I suspect you may have the best success by using monocular depth approaches (for example something like this: https://github.com/nianticlabs/monodepth2).

After that I have to do the same for Python I think, and then I have to find out how to figure out how to use a library like https://github.com/ialhashim/DenseDepth or https://github.com/nianticlabs/monodepth2 for that GStreamer plugin (or element, still trying to grasp the terminology here)

I would go with https://github.com/TRI-ML/packnet-sfm. There is plenty of support from the community on the Github repo, it is very well known and very well tested. You can see it as a modern version of MonoDepth 2.