Mean generalization capacity across all identity-preserving transformations (Fig. 7d). Included are mean generalization capacity estimates based on the performance of: the linear classifier analysis including the neurons recorded from both subjects and when spikes were counted in a 218 ms window (the duration of each stimulus), the linear classifier performance for subjects 1 and 2 individually, the correlation-based classifier, the linear classifier analysis when spikes were counted in 25, 50, and 100 ms windows, and the linear classifier analysis when mean firing rates were computed across trials on which eye position deviated <0.2° and trial-to-trial variability was simulated with a Poisson process (see Results). In 25 ms windows, performance of the V4 population on some of the images (Fig. 7b) was not significantly different from chance and thus generalization could not be assessed.