Intrinsic Image Popularity Assessment
share
The goal of research in image popularity assessment (IPA) is to develop computational models that can automatically predict the potential of a social image being popular over the Internet. Here, we aim to single out the contribution of visual content to image popularity, i.e., intrinsic image popularity that is of great practical importance. Specifically, we first describe a probabilistic method to generate massive popularity-discriminable image pairs, based on which the first large-scale database for intrinsic IPA (I^2PA) is established. We then learn computational models for I^2PA by optimizing deep neural networks for ranking consistency with millions of popularity-discriminable image pairs. Experiments on Instagram and other social platforms demonstrate that the optimized model outperforms state-of-the-art methods and humans, and exhibits reasonable generalizability. Moreover, we conduct a psychophysical experiment to analyze various aspects of human behavior in I^2PA.
READ FULL TEXT