Generating Gold Questions for Difficult Visual Recognition Tasks
Diane Larlus, Florent Perronnin, Pramod Kompalli, Vivek Mishra
Gold questions are a standard mechanism to detect insincere
workers on crowdsourcing platforms. They usually
rely on the assumption that workers should obtain perfect
accuracy on the task. In this work, we are interested in
crowdsourcing difficult multi-class visual recognition tasks,
for which this assumption is not met, and we propose a
novel method for generating gold questions in this context.
Workshop on Computer Vision and Human Computation at the CVPR Conference in Columbus, Ohio, June 23-28, 2014.
2014-019.pdf (986.02 kB)