Modelfest Index

Welcome to the Modelfest - image database and threshold data collection group.

If you wish to participate in this project please contact thom@neurometrics.com.

We have finished gathering year-one static 2D spatial image threshold data. Below is an abstract from the 1999 SPIE meeting which summarized our plans. We are now beginning data collection plans for static spatio-chromatic image and spatio-temporal image sequences.

The development of an image/threshold database for designing and testing human vision models.

Thom Carney(a&b), Stanley A. Klein(b), Christopher W. Tyler(c), Amnon D. Silverstein(d),
Brent Beutter(e), Dennis Levi(f), Andrew B. Watson(e), Adam J. Reeves(g),
Anthony M. Norcia(c), Chien-Chung Chen(c), Walter Makous(h) and Miguel P. Eckstein(i)

(a) Neurometrics Institute, 2400 Bancroft Way, Berkeley CA, 94704
(b) School of Optometry, University of California at Berkeley, Berkeley, CA, 94720
(c) Smith-Kettlewell Eye Research Institute, San Francisco, CA, 94115
(d) Hewlett-Packard Inc., HP Labs, Palo Alto, CA.
(e) NASA Ames Research Center, Moffett Field, CA, 94035-1000
(f) College of Optometry, University of Houston, Houston, TX, 77204-6052
(g) Department of Psychology, Northeastern University, Boston, MA, 02115
(h) Center for Visual Science, University of Rochester, Rocherter, NY, 14627-0270
(i) Medical Physics & Imaging, Cedars Sinai Medical Center, Los Angles, CA, 90048-1864

Abstract:

Models that predict human performance on narrow classes of stimuli abound in the vision science literature. The image compression community needs robust general-purpose computational human visual system (HVS) models to evaluate image fidelity/quality. Psychophysical measures of image quality are too costly: and time consuming to evaluate the impact each compression algorithm modification might have on image quality.
Several general-purpose early HVS models currently exist, but direct models comparisons on the same data sets are rarely made. Moreover, researchers designing a new model are confronted with the decision of what data set to use for normalizing mechanisms. To solve some of these issues about 40 researchers interested in vision modeling problems have formed a group tentatively called Modelfest. Some members of the group are developing a database of test images with threshold data that will be posted on the WEB for all modellers to use in HVS model design and testing. We also hope to include threshold data and stimulus specifications that are available in the literature which provide critical information for model design and testing. The space of possible stimuli is enormous; for this first year's effort of data collection we have decided to limit our goals to detection thresholds for static gray scale 2D images. In future years, the database will be extended to include discrimination (masking) as well as detection thresholds for dynamic color and gray scale image sequences. The purpose of this presentation is to invite the Electronic Imaging community to participate in this effort and inform them of the developing data set, which will be available to all interested researchers.
Issues to be discussed include: 1) what hardware constraints need be considered in selecting the detection stimulus set, 2) what psychophysical methods will be used for data collection, 3) what are the 2D stimulus patterns that best test aspects of HVS models, and finally, 4) how to ensure that the stimuli and data are easily accessed by researchers irrespective of their hardware platform, modeling environment or operating system.