Fig. 3
From: Attention and stimulus structure interact during ensemble encoding of facial expression
Joint ensemble and single-face space along with examples of behavior-based image reconstructions from Experiment 1. Ensemble representations were projected in an expression face space and reconstructed from its structure, separately for each facial identity (left: ID1; right: ID2). (A) Both single faces and ensembles evinced a marked separation by valence in face space as well as clear within-valence variability: plots show the first two dimensions and percent variance explained (PVE) of a joint single-ensemble face space averaged across participants for visualization purposes (each ensemble, marked by red and green circles, consists of 6 of the 7 within-valence expressions, labelled based on the absent expression; pos positive-expression ensemble, neg negative-expression ensemble). (B) Image reconstructions capture visual representations of ensembles (1 and 2) and single-face stimuli (3 and 4)—ensemble stimuli are replaced in the upper row by the pixelwise average of their individual faces. Examples of positive and negative expressions are displayed along with their reconstruction accuracy (averaged across participants).
