Appendix 2—figure 6. Cueing spatial attention has little effect on performance.

(A) Covert spatial attention was cued to the area of the largest difference between the images (70% of trials; half from conv5 feature MSE; half from pixel MSE) via a wedge stimulus presented before the trial. On 15% of trials the wedge cued an invalid location (smallest pixel MSE), and on 15% of trials no cue was provided (circle stimulus). (B) Performance as a function of cueing condition for 30 participants. Points show grand mean (error bars show $\pm 2$ SE), lines link the mean performance of each observer for each pooling model (based on at least 30 trials; median 65). Blue lines and shaded area show the population mean estimate and 95% credible intervals from the mixed-effects model. Triangle in the Uncued condition replots the average performance from CNN 8 in Figure 3 for comparison. Images from the MIT 1003 dataset (Judd et al., 2009, https://people.csail.mit.edu/tjudd/WherePeopleLook/index.html) and reproduced under a CC-BY license (https://creativecommons.org/licenses/by/3.0/) with changes as described in the Materials and methods.