Introduction • Goal - Automatically annotate segments in weakly labeled video taken from YouTube

• Challenges - Learning from weakly labeled data - Handling label noise in YouTube tags - Parallelize to deploy over large amounts of YouTube data Our Problem Setup

Our Algorithm: CRANE • Input: uncertain positive segments, large set of negative segments • Output: ranked positive segments by probability of belonging to our concept

Intuition: Positive segments are less likely to belong to our concept if they are near many negative segments. Sample Object Segmentations

Inductive Segment Annotation [top two rows] Transductive Segment Annotation [bottom two rows]

Common Failure Cases [1] P. Siva, C. Russell, and T. Xiang. In defence of negative mining for annotating weakly labelled data. ECCV 2012. [2] M. Grundmann, V. Kwatra, M. Han, and I. Essa. Efficient hierarchical graph-based video segmentation. CVPR 2010. [3] G. Hartmann et al. Weakly supervised learning of object segmentations from web-scale video. ECCV 2012 Workshop. [4] A. Prest et al. Learning object class detectors from weakly annotated video. CVPR 2012. Quantitative Results Transductive Segment Annotation (annotating a dataset)

• Inductive Segment Annotation (novel object segmentation)