K. Tang, R. Sukthankar, J. Yagnik, L. Fei-Fei. Discriminative Segment Annotation in Weakly Labeled Video. Conference on Computer Vision and Pattern Recognition, 2013.
This paper proposes CRANE - Concept Ranking According to Negative Exemplars - for semantic two-class segmentation of weakly labeled videos. The task can be summarized as follows: Given an oversegmentation of a video, tagged weakly with a concept such as "cat" or "dog", decide which segments actually belong to the concept. As known from semantic segmentation of weakly-labeled images, common difficulties are high in-class variation of concepts like "cat" and "dog" as well as the unknown location of the concept. In videos, an additional difficulty is the unknown temporal location of the concept.
This paper proposes CRANE - Concept Ranking According to Negative Exemplars - for semantic two-class segmentation of weakly labeled videos. The task can be summarized as follows: Given an oversegmentation of a video, tagged weakly with a concept such as "cat" or "dog", decide which segments actually belong to the concept. As known from semantic segmentation of weakly-labeled images, common difficulties are high in-class variation of concepts like "cat" and "dog" as well as the unknown location of the concept. In videos, an additional difficulty is the unknown temporal location of the concept.
An overview of CRANE is given on the paper's webpage.