The key difference from traditional MRF based image segmentation methods is that we infer semantic segments in the label space instead of the pixel space.
We define semantic segments as short self-explanatory video segments that can be used as building blocks for the highlights video.
To make the evaluation feasible, we selected 100 pitching scenes (semantic segments) of about 15 s in duration from a baseball game.
Besides the typical semantic segmentation approaches, many approaches aim at the accuracy improvement of object segment, such as the interactive segmentation [16 18] and the co-segmentation [19 21].
We observe that audiovisual salient segments correlate mostly with Token-Token, Metaphor, and Metonymy relations, audio salient segments correlate mostly with Token-Token, Metonymy, and Type-Token, visual salient segments correlate with Metaphor and Type-Token, while semantic salient segments with Token-Token and Metaphor.
Within KCAR, we explore the semantics underlying each sensor event and use semantic dissimilarity to segment a continuous sensor sequence into fragments, each of which corresponds to one ongoing activity.
Previous works focus on assigning a unique category label to each pixel correctly, generating region segments with semantic information.
In these works, the prior of latent object is provided with either the user scribbles or the object coherency to generate exact segments, while these segments usually have no semantic information of objects.
Here, we focus on coherence specifically and treat it as a level of semantic relationship among specific text segments.
This significant user-experience problem motivates our new development of extending user interactions from semantic zooming to marking interesting segments on the curve.
