These scripts underlie an in-progress human behavioral experiment being conducted by Katherine Hermann (Stanford University) and Chaz Firestone (Johns Hopkins University).
-
To enable running experiment, add a directory named
jsPsych
containing jsPsych==6.3.0 files (available athttps://www.jspsych.org
[1]) toexperiment/js/
. -
To enable running tests (written in Jest), navigate to
generate_experiment_specs
, install Node and npm, and runnpm install
. To run tests:npm test
. -
To support running the experiment with the "gst" images, copy the stimuli from
https://github.com/rgeirhos/texture-vs-shape
(do not preserve the subdirectory structure). -
Gst scramble masks are available upon request.
The main experiment script (experiment/main.html
) supports 2 stimulus sets and 3 mask types, for a total of 6 possible experiments.
-
Set the
stimCondition
to "gst" (Geirhos Style Transfer images) or "baker" (Baker images). -
Set the
maskCondition
to "pinknoise", "scramble" (scramble masks; will use scrambles corresponding to thestimCondition
), or "nomask" (shows a red fixation cross during the response period). -
If running as experiment, replace
redirectUrl
with a valid Prolific study url.
Baker images and scramble masks were generated from materials provided by Nicholas Baker [2], and are included here by permission of authors.
Geirhos Style Transfer images are from [3].
[1] De Leeuw, J. R. (2015). jsPsych: A JavaScript library for creating behavioral experiments in a Web browser. Behavior research methods, 47(1), 1-12.
[2] Baker, N., Lu, H., Erlikhman, G., & Kellman, P. J. (2018). Deep convolutional networks do not classify based on global object shape. PLoS computational biology, 14(12), e1006613.
[3] Geirhos, R., Rubisch, P., Michaelis, C., Bethge, M., Wichmann, F. A., & Brendel, W. (2018). ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness. arXiv preprint arXiv:1811.12231.