Fig. 4

Human psychophysics experiments suggest that object position discrimination is unaffected by stimulus background variation. Since object position representation in V4 neurons is robust to background variation, we tested whether changing background properties affects the decoding of object position in humans. (a) Human psychophysical task. Two images containing a object were presented, with two masks in between. Participants were instructed to report whether the relative position of the object in image 2 was left or right of that in image 1. In blocks, background rotation and/or depth were held constant or varied as described below. (b) Averaged (across participants, N = 10) psychometric functions for each background variation condition (individual participant performance shown in Supp. Fig. 5). Three background conditions were tested: black: no background variation between the two presentations of the object on each trial; red: background rotation changed randomly across the two presentations of the object, but background depth was held fixed; blue: both background rotation and depth were randomized across the two presentations of the object. Across participants, object position change detection performance was similar across the three background variation conditions (paired t-tests on thresholds between each pair, p > 0.05, Bonferroni corrected). Bayes Factor Analysis for the main effect of background variations suggests strong evidence for the absence of the effect of background variation in choices (1/BF = 34.27; see Methods for details). This value of 1/BF provides support for the absence of a background effect in the human psychophysics. (c) To compare human behavior and monkey electrophysiology, we selected stimulus presentations in the monkey experiments to approximate the three background variation levels used in the human psychophysics experiment (see Methods for details of sample matching). We trained linear discriminants to separate trials into right or left position shift, for each background variation condition. During training, the trials with the object in the central position were randomly assigned to be left or right. Classifier performance also did not differ substantially across the background variation conditions (paired t-tests on thresholds between each pair, p > 0.05, Bonferroni corrected). Similar to (b) above, Bayes Factor analysis suggests strong evidence for the absence of the effect of background (1/BF = 26.25 for monkey 1 and 1/BF = 8.99 for monkey 2). These values of 1/BF provide support for the absence of a background effect of a background effect in the neural decoding.