Video Memorability Prediction via Image-Based Deep Neural Net Models
Abstract
Abstract
The memorability score represents the probability of a general audience remembering that media. Memorability as an attribute of a video is useful for advertisement, content design (media sorting for photographers), and education. Higher memorability correlates with greater popularity. The results of our study show that more video frames do not necessarily improve the prediction performance and that ResNet output as a feature alone is highly predictive for the video mermorability prediction task, to the point of plausibly overshadowing the inclusion of AMNet and caption features in our model.
Description
Department of Computer Science
Rights
File access restricted due to FERPA regulations