Visual lifelogs tend to grow in size and the visual content presented therein is very complex. This complexity is hard to overcome in application such as localization of the camera wearer using only visual information. More clues about particular place can be gained from previously constructed 3D models of several salient places that can be found indoors. In [PDF] it was shown that using constructed Structure from Motion (SfM) models for some places can be perform better than standard 2D approach with global image descriptor classification.