Image Classification and Retrieval from User-Supplied Tags
This addresses the high cost of manual annotation for image classification, benefiting researchers and practitioners in computer vision by leveraging freely available online data.
The paper tackles the problem of image classification by learning directly from noisy user-supplied tags, such as those from Flickr, without costly manual filtering. It shows that this approach can achieve similar or superior results compared to databases with expensive manual annotations.
This paper proposes direct learning of image classification from user-supplied tags, without filtering. Each tag is supplied by the user who shared the image online. Enormous numbers of these tags are freely available online, and they give insight about the image categories important to users and to image classification. Our approach is complementary to the conventional approach of manual annotation, which is extremely costly. We analyze of the Flickr 100 Million Image dataset, making several useful observations about the statistics of these tags. We introduce a large-scale robust classification algorithm, in order to handle the inherent noise in these tags, and a calibration procedure to better predict objective annotations. We show that freely available, user-supplied tags can obtain similar or superior results to large databases of costly manual annotations.