Joint people, event, and location recognition in personal photo collections using cross-domain context

摘要: We present a framework for vision-assisted tagging of personal photo collections using context. Whereas previous efforts mainly focus on people, we develop unified approach to jointly tag across multiple domains (specifically events, and locations). The heart our is generic probabilistic model context that couples the through set cross-domain relations. Each relation models how likely instances in two are co-occur. Based this model, derive an algorithm simultaneously estimates relations infers unknown tags semi-supervised manner. conducted experiments well-known datasets obtained significant performance improvements both people location recognition. also demonstrated ability infer event labels with missing timestamps (i.e. no features).