Is There a Way to Train a System to Look Around a Scene?

One place that i think there could be a lot to learn from kono psychology is in sort of perception. Because when humans are looking at a scene, we don't like stare at the centre and then process the whole scene. We're glancing all round it. There's been a lot of worknike what parts of scenes people attend to and whatever they may be getting out of that. Could potentially be a lot of value to be gotten in training systems to actually look around an image as they're processing it. And so one would expect it to be much more efficient to be able to do this correctly for vidior.