Fig. 3From: Efficient video collection association using geometry-aware Bag-of-Iconics representationsAlignment pipeline overviewBack to article page