CHDH Seminar Series 2020: Modelling visual search using features from deep convolutional neural networks — Kris Ehinger
Presenter: Dr Krista Ehinger
Date: Monday 28 September
Time: 12 - 1 PM
Location: Via Zoom. Please use this link at the time of the event to join: https://unimelb.zoom.us/j/96377607982?pwd=NDVDaUQrQStqVXlyZmp5SEhoMExFQT09
When people search for objects in images, their attention is guided by features of the search target, such as its colour or shape. Various models have been proposed to explain this “pre-attentive” feature guidance (e.g., Wolfe, 2013). However, these models generally focus on simple features such as colour, brightness, or line orientation because it is easy to understand the similarity between a search target and distractors along these dimensions. It is difficult to extend these models to search for more complex visual features like shape, or to search for objects in natural images in general, because this requires a model of pre-attentive processing that can capture the similarity between a search target and distractors along many different feature dimensions. Features learned by “deep” convolutional neural networks (CNNs) may be useful as a proxy for this pre-attentive feature space. Kris will show how these features can be used to model more complex visual search tasks, such as search for targets defined by shape, and “hybrid” visual-memory search, in which observers must decide if any object in a visual display is a member of a memorized target set. The results suggest that the visual representation learned by a “deep” CNN is a reasonable approximation of early or pre-attentive feature processing in the human visual system. CNN features can be incorporated into cognitive models to better understand human performance in complex visual tasks.
Krista Ehinger is a Senior Lecturer in the School of Computing and Information Systems at the University of Melbourne. She received her Ph.D. in the Department of Brain and Cognitive Sciences at the Massachusetts Institute of Technology in 2013. She was a postdoc at Harvard Medical School from 2013-2016 and a postdoc at York University from 2016-2019. Her work focusses on the intersection of human and computer vision for tasks such as scene recognition, visual search and depth perception in natural scenes.