TBH that's better than I expected. I'd call the column masking "barely adequate" rather than "correctly" but it's at least making an attempt to reconstruct the scene and do additive composition in realtime.

Wasn't the real breakthrough meant to be to do with variable focus lightfield voodoo meaning that the rendered components looked like they 'fitted in' with the scene and could occlude as well as be occluded, rather than being additive CGI floating in front of it?