Yes they make one using simple models that roughly fit the geometry on screen. Then transform for parallax. That may reveal hidden areas so they fill them in either using an algorithm from surrounding areas and recent frames or manually.