Abstract

We introduce tools and methodologies to collect high quality, large scale fifine-grained computer vision datasets using citizen scientists – crowd annotators who are passionate and knowledgeable about specifific domains such as birds or airplanes. We worked with citizen scientists and domain experts to collect NABirds, a new high quality dataset containing 48,562 images of North American birds with 555 categories, part annotations and bounding boxes. We fifind that citizen scientists are signifificantly more accurate than Mechanical Turkers at zero cost. We worked with bird experts to measure the quality of popular datasets like CUB- 200-2011 and ImageNet and found class label error rates of at least 4%. Nevertheless, we found that learning algorithms are surprisingly robust to annotation errors and this level of training data corruption can lead to an acceptably small increase in test error if the training set has suffificient size. At the same time, we found that an expert-curated high quality test set like NABirds is necessary to accurately measure the performance of fifine-grained computer vision systems. We used NABirds to train a publicly available bird recognition service deployed on the web site of the Cornell Lab of Ornithology.1