Also SFM matching between sequences needs to be improved for this to work properly. For example, these sequences already have decent individual reconstructions, but there are three different reconstructions of the same thing: Point clouds, are they coming back? - #16 by enteq (also see other threads linked there)