|Description (include details on usage, files and paper references)||We present a dataset to address the problem of visual privacy - where users unintentionally leak private information when sharing personal images online, such as on Flickr, Twitter or Facebook. First, we categorize "private information" into 68 privacy attributes (e.g., religion, face, sexual orientation, passport, etc.). Second, we run user studies to understand privacy preferences of users over these attributes.
This dataset includes:
a) 22k images annotated with 68 privacy attributes
b) User study of 305 participants indicating their privacy preferences over the 68 privacy attributes
c) User study of 50 participants containing for around 300 images i) their privacy preference for an attribute ii) their evaluation of privacy risk from the image for that attribute
This dataset can be used for:
a) Multi-label classification
b) Predicting a Privacy Risk Score from images given a users privacy preference
Tribhuvanesh Orekondy, Bernt Schiele, Mario Fritz, "Towards a Visual Privacy Advisor: Understanding and Predicting Privacy Risks in Images", in IEEE International Conference on Computer Vision, 2017.