The rule of being specific also matches this title. This is for very very limited artificial video game footage and a database of known 3D scenes from the same. Hardly what I expected.
They are using recorded footage and depth maps from a FIFA game to 're-match' new footage with depth.
While they are using "artificial video game footage," as the basis for their process, they are applying it to real soccer game footage. Limited, yes, but quite clever in its simplicity. (Obviously, in general, the idea of using simulated data as a training set for supervised machine learning isn't new, but who would have thought that an off-the-shelf videogame would be realistic enough to be used to reconstruct 3d scenes from actual soccer footage.)