Achievement

IGERT fellows Matthew Tong and Christopher Kanan in collaboration with Lingyun Zhang and Professor Gary Cottrell (IGERT co-PI) have extended the computer implementation of the SUN model of salience (previously also worked on by IGERT fellow Nick Butko, IGERT alumnus Tim Marks, IGERT faculty Javier Movellan, graduate student Honghao Shan together with Zhang, Tong, and Cottrell) to include task information about the appearance of targets and their likely locations. The resulting implementation achieves state of the art performance in predicting human fixations doing a counting task in natural scenes (e.g. counting the number of cups in a kitchen scene) while shedding some light on the role of target appearance and context in guiding overt attention. The model even made some of the same mistakes that humans make. This work is an example of how research from human vision can improve machine vision algorithms. The model was recently published in the journal Visual Cognition.