Embracing Error to Enable Rapid Crowdsourcing
Based on research published in this paper.
Our goal will be to separate a set of dog images from the images on the right. When you click play, a random set of images will be flashed in front of you. Press the spacebar whenever you see a dog. Mobile users can click anywhere on the screen when the slideshow starts.
Current Precision @ : , Recall @ :
Don't worry about reacting late, our model accounts for worker delays and will attempt to pick out the correct dog images. Every time you do the task, the model will move the most likely dog images to the top. The model will also print precision and recall curves at the bottom so you can track your progress.
Every time you do the task, the scores for the images are updated and a new precision and recall curve is drawn. So far, you have done tasks.