Timezone: »

 
Simulating Human Gaze with Neural Visual Attention
Leo Schwinn · Doina Precup · Bjoern Eskofier · Dario Zanca

Sat Dec 03 10:30 AM -- 11:30 AM (PST) @
Event URL: https://openreview.net/forum?id=bai3OCRQF_N »

Existing models of human visual attention are generally unable to incorporate direct task guidance and therefore cannot model an intent or goal when exploring a scene. To integrate guidance of any downstream visual task into attention modeling, we propose the Neural Visual Attention (NeVA) algorithm. To this end, we impose to neural networks the biological constraint of foveated vision and train an attention mechanism to generate visual explorations that maximize the performance with respect to the downstream task. We observe that biologically constrained neural networks generate human-like scanpaths without being trained for this objective. Extensive experiments on three common benchmark datasets show that our method outperforms state-of-the-art unsupervised human attention models in generating human-like scanpaths.

Author Information

Leo Schwinn (Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU))
Doina Precup (McGill University / Mila / DeepMind Montreal)
Bjoern Eskofier (Friedrich-Alexander University Erlangen-Nuernberg)
Dario Zanca (FAU Erlangen-Nürnberg)
Dario Zanca

Dario Zanca received the B.Sc. and M.Sc. Degrees in Mathematics from the University of Palermo, Italy. He received the Ph.D. in Smart Computing from the University of Florence, Italy, working partly at the California Institute of Technology (Caltech), Pasadena, United States. He worked as postdoc researcher at the Department of Medicine, Surgery and Neuroscience at the University of Siena, Italy. He is currently postdoc researcher at the Department Artificial Intelligence in Biomedical Engineering, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany. His research interests broadly fall into the areas of computer vision and machine learning with emphasis on computational biologically-inspired self-learning systems, human attention, and eye-tracking.

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors