Beyond detection of cows: Semantic and visual analysis of Flickr photos

http://aksw.org/Talk/BeyondDetectionOfCows an entity of type: Talk

Vortrag
xsd:string
people:Events/2013/LeipzigerSemanticWebTag/Menu
Category-level image classification describes the process of automatically assigning a label to an image by means of computer vision and machine learning techniques. Current approaches have significantly advanced and render image classification ready for practical use. Despite all progress that has been made, most approaches heavily rely on manual curation of training data for learning the model of a specific category – a very time-consuming process, which needs to be reiterated whenever extending the classifier to an additional category. Therefore, the cognitive repertoire of automatic classifiers is typically rather low. We discuss approaches for automatic mining of Flickr photos for collecting training data. The visual variance in Flickr images can be considered as large, covering a plethora of different categories. Moreover, many Flickr photos are provided with user generated annotations such as tags and titles. We seek to alleviate the process of training data generation by mining Flickr for images representing a specific category based on these annotations. However, annotations must be considered as noisy since they do not necessarily label the depicted objects or scene. We therefore consider approaches to semantically analyze annotations and visual image content in order to limit training images to those reliably depicting the respective category.
xsd:string
<http://www.hpi.uni-potsdam.de/meinel/lehrstuhl/team_fotos/current_phd_students/christian_hentschel.html#id>
↪ Christian Hentschel
2013-09-23
xsd:date
people:Events/2013/LeipzigerSemanticWebTag
↪ 5. Leipziger Semantic Web Tag (LSWT2013)
aksw:Talk
↪ Vortrag
Beyond detection of cows: Semantic and visual analysis of Flickr photos
xsd:string

inverse relations

1.29919ms